diff --git a/packages/agents/content/_partials/frontmatter-via-script.md b/packages/agents/content/_partials/frontmatter-via-script.md
deleted file mode 100644
index 5155a758..00000000
--- a/packages/agents/content/_partials/frontmatter-via-script.md
+++ /dev/null
@@ -1,6 +0,0 @@
-Run `resolve-frontmatter.sh` via the Bash tool. It emits a JSON object with the universal artifact fields (`branch`, `commit`, `baseSha`, `pr`, `ticket_id`, `ticket_ref`, `platform`, `timestamp`, `run_id`). Use those values verbatim for the matching YAML keys. Optional fields the script omits from its output (`baseSha`, `pr`, `ticket_id`, `ticket_ref`, `run_id`) must be omitted from the frontmatter too — do not emit `null` or empty strings.
-
-If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
-
-Set these skill-specific values inline (not in the script's output):
-<!-- children -->
diff --git a/packages/agents/content/scripts/__tests__/resolve_frontmatter_test.sh b/packages/agents/content/scripts/__tests__/resolve_frontmatter_test.sh
index 7de4bb4c..3f79dce2 100644
--- a/packages/agents/content/scripts/__tests__/resolve_frontmatter_test.sh
+++ b/packages/agents/content/scripts/__tests__/resolve_frontmatter_test.sh
@@ -75,6 +75,26 @@ It "emits run_id when present"
 When call emit_json "main" "abc1234" "" "" "" "" "github" "2026-05-16T00:00:00Z" "20260516-143946Z"
 The output should include '"run_id": "20260516-143946Z"'
 End
+
+It "emits a fully-populated argument set in canonical key order"
+expected_json() {
+  cat <<'JSON'
+{
+  "branch": "main",
+  "commit": "abc1234",
+  "platform": "github",
+  "timestamp": "2026-05-16T00:00:00Z",
+  "baseSha": "deadbee",
+  "pr": "https://github.com/x/y/pull/1",
+  "ticket_id": "537",
+  "ticket_ref": "#537",
+  "run_id": "20260516-143946Z"
+}
+JSON
+}
+When call emit_json "main" "abc1234" "deadbee" "https://github.com/x/y/pull/1" "537" "#537" "github" "2026-05-16T00:00:00Z" "20260516-143946Z"
+The output should equal "$(expected_json)"
+End
 End
 
 Describe "warn_pr_failure"
@@ -143,3 +163,542 @@ When call resolve_base_sha "HEAD"
 The output should equal "$ref"
 End
 End
+
+Describe "needs_yaml_quoting"
+It "returns true for empty values"
+When call needs_yaml_quoting ""
+The status should be success
+End
+
+It "returns false for plain alphanumerics"
+When call needs_yaml_quoting "foo123"
+The status should be failure
+End
+
+It "returns true for values where a colon is followed by whitespace"
+When call needs_yaml_quoting "key: value"
+The status should be success
+End
+
+It "returns false for values where a colon is followed by a non-space character"
+When call needs_yaml_quoting "key:value"
+The status should be failure
+End
+
+It "returns true for values ending in a trailing colon"
+When call needs_yaml_quoting "trailing:"
+The status should be success
+End
+
+It "returns true for values containing pound sign"
+When call needs_yaml_quoting "#537"
+The status should be success
+End
+
+It "returns true for values with leading whitespace"
+When call needs_yaml_quoting " leading"
+The status should be success
+End
+
+It "returns true for values with trailing whitespace"
+When call needs_yaml_quoting "trailing "
+The status should be success
+End
+
+It "returns true for values beginning with a hyphen"
+When call needs_yaml_quoting "-leading"
+The status should be success
+End
+
+It "returns true for values beginning with a question mark"
+When call needs_yaml_quoting "?leading"
+The status should be success
+End
+
+It "returns true for values beginning with a colon"
+When call needs_yaml_quoting ":leading"
+The status should be success
+End
+
+It "returns false for URLs (no special chars under predicate)"
+When call needs_yaml_quoting "https://github.com/x/y/pull/1"
+The status should be failure
+End
+
+It "returns true for values containing brackets"
+When call needs_yaml_quoting "a[b]c"
+The status should be success
+End
+
+It "returns true for values containing braces"
+When call needs_yaml_quoting "{key}"
+The status should be success
+End
+
+It "returns true for values containing commas"
+When call needs_yaml_quoting "a,b"
+The status should be success
+End
+
+It "returns true for values containing pipe"
+When call needs_yaml_quoting "a|b"
+The status should be success
+End
+
+It "returns true for values containing backtick"
+When call needs_yaml_quoting "a\`b"
+The status should be success
+End
+
+It "returns true for values containing an asterisk"
+When call needs_yaml_quoting "a*b"
+The status should be success
+End
+
+It "returns true for values containing an ampersand"
+When call needs_yaml_quoting "a&b"
+The status should be success
+End
+
+It "returns true for values containing an exclamation mark"
+When call needs_yaml_quoting "a!b"
+The status should be success
+End
+
+It "returns true for values containing a greater-than sign"
+When call needs_yaml_quoting "a>b"
+The status should be success
+End
+
+It "returns true for values containing a less-than sign"
+When call needs_yaml_quoting "a<b"
+The status should be success
+End
+
+It "returns true for values containing a percent sign"
+When call needs_yaml_quoting "a%b"
+The status should be success
+End
+
+It "returns true for values containing an at sign"
+When call needs_yaml_quoting "a@b"
+The status should be success
+End
+
+It "returns true for values containing a double quote"
+When call needs_yaml_quoting 'a"b'
+The status should be success
+End
+End
+
+Describe "yaml_quote"
+It "leaves bare values unquoted"
+When call yaml_quote "foo123"
+The output should equal "foo123"
+End
+
+It "wraps unsafe values in single quotes"
+When call yaml_quote "#537"
+The output should equal "'#537'"
+End
+
+It "doubles embedded single quotes inside the wrapper"
+When call yaml_quote "it's"
+The output should equal "'it''s'"
+End
+
+It "quotes empty values as empty string"
+When call yaml_quote ""
+The output should equal "''"
+End
+
+It "leaves URLs unquoted"
+When call yaml_quote "https://github.com/x/y/pull/1"
+The output should equal "https://github.com/x/y/pull/1"
+End
+End
+
+Describe "emit_yaml_flow_list"
+It "emits an empty flow list for empty values"
+When call emit_yaml_flow_list "items" ""
+The output should equal "items: []"
+End
+
+It "emits single-element flow lists in bracket form"
+When call emit_yaml_flow_list "commits" "a1b2c3d"
+The output should equal "commits: [a1b2c3d]"
+End
+
+It "splits comma-separated values into list elements"
+When call emit_yaml_flow_list "commits" "a1b2c3d,e4f5g6h"
+The output should equal "commits: [a1b2c3d, e4f5g6h]"
+End
+
+It "auto-quotes elements that contain unsafe glyphs"
+When call emit_yaml_flow_list "refs" "main,#537"
+The output should equal "refs: [main, '#537']"
+End
+End
+
+Describe "add_extra"
+add_extra_setup() {
+  unset extra_keys extra_values extra_kinds
+  declare -ga extra_keys=()
+  declare -gA extra_values=()
+  declare -gA extra_kinds=()
+}
+
+BeforeEach "add_extra_setup"
+
+It "records insertion order across mixed extra kinds"
+add_extra "scalar" "title=hello" extra_keys extra_values extra_kinds
+add_extra "list" "commits=a,b" extra_keys extra_values extra_kinds
+add_extra "scalar" "scope=root" extra_keys extra_values extra_kinds
+When call test "${#extra_keys[@]}" -eq 3
+The status should be success
+End
+
+It "splits the key on the first equals sign"
+add_extra "scalar" "title=a=b=c" extra_keys extra_values extra_kinds
+When call echo "${extra_values[title]}"
+The output should equal "a=b=c"
+End
+
+It "tracks the kind per key"
+add_extra "list" "commits=a,b" extra_keys extra_values extra_kinds
+When call echo "${extra_kinds[commits]}"
+The output should equal "list"
+End
+
+It "fails when the argument has no equals sign"
+When run add_extra "scalar" "bad_arg" extra_keys extra_values extra_kinds
+The status should be failure
+The stderr should include "missing '='"
+End
+
+It "fails when the key is empty"
+When run add_extra "scalar" "=value" extra_keys extra_values extra_kinds
+The status should be failure
+The stderr should include "empty key"
+End
+
+It "warns and overwrites when the same key is added twice"
+add_extra "scalar" "key=first" extra_keys extra_values extra_kinds
+add_extra_again() {
+  add_extra "scalar" "key=second" extra_keys extra_values extra_kinds
+  # Emit observable state on stdout for the assertion to inspect.
+  echo "value=${extra_values[key]}"
+  echo "count=${#extra_keys[@]}"
+}
+When call add_extra_again
+The stderr should include "duplicate"
+The output should include "value=second"
+The output should include "count=1"
+End
+End
+
+Describe "apply_override"
+apply_override_setup() {
+  unset overrides
+  declare -gA overrides=()
+}
+
+BeforeEach "apply_override_setup"
+
+It "returns the resolved value when no override is registered"
+When call apply_override "branch" "main" overrides
+The output should equal "main"
+End
+
+It "returns the override when one is registered"
+overrides[branch]="custom-branch"
+When call apply_override "branch" "main" overrides
+The output should equal "custom-branch"
+End
+
+It "force-omits when the override value is empty"
+overrides[run_id]=""
+When call apply_override "run_id" "20260516Z" overrides
+The output should equal ""
+End
+End
+
+Describe "emit_yaml"
+emit_yaml_setup() {
+  unset yaml_keys yaml_values yaml_kinds
+  declare -ga yaml_keys=()
+  declare -gA yaml_values=()
+  declare -gA yaml_kinds=()
+}
+
+BeforeEach "emit_yaml_setup"
+
+It "wraps the output in --- delimiters"
+When call emit_yaml \
+  "create-devlog" "2026-05-16T00:00:00Z" "deadbee" "true" "" \
+  "" "" "main" "abc1234" "" "" \
+  yaml_keys yaml_values yaml_kinds
+The line 1 of output should equal "---"
+The output should end with "---"
+End
+
+It "emits provenance block in canonical order"
+When call emit_yaml \
+  "create-devlog" "2026-05-16T00:00:00Z" "deadbee" "true" "claude-opus" \
+  "" "" "main" "abc1234" "" "" \
+  yaml_keys yaml_values yaml_kinds
+The output should include "provenance:"
+The output should include "skill: create-devlog"
+The output should include "timestamp: 2026-05-16T00:00:00Z"
+The output should include "baseSha: deadbee"
+The output should include "isInteractive: true"
+The output should include "model: claude-opus"
+End
+
+It "omits provenance.baseSha when empty"
+When call emit_yaml \
+  "skill-x" "2026-05-16T00:00:00Z" "" "false" "" \
+  "" "" "main" "abc1234" "" "" \
+  yaml_keys yaml_values yaml_kinds
+The output should not include "baseSha"
+End
+
+It "omits provenance.model when empty"
+When call emit_yaml \
+  "skill-x" "2026-05-16T00:00:00Z" "deadbee" "false" "" \
+  "" "" "main" "abc1234" "" "" \
+  yaml_keys yaml_values yaml_kinds
+The output should not include "model"
+End
+
+It "emits isInteractive as a bare boolean"
+When call emit_yaml \
+  "skill-x" "2026-05-16T00:00:00Z" "deadbee" "false" "" \
+  "" "" "main" "abc1234" "" "" \
+  yaml_keys yaml_values yaml_kinds
+The output should include "isInteractive: false"
+The output should not include "isInteractive: 'false'"
+End
+
+It "emits canonical top-level fields after provenance"
+When call emit_yaml \
+  "skill-x" "2026-05-16T00:00:00Z" "deadbee" "false" "" \
+  "537" "#537" "main" "abc1234" "https://github.com/x/y/pull/1" "20260516-143946Z" \
+  yaml_keys yaml_values yaml_kinds
+The output should include "ticket_id: 537"
+The output should include "ticket_ref: '#537'"
+The output should include "branch: main"
+The output should include "commit: abc1234"
+The output should include "pr: https://github.com/x/y/pull/1"
+The output should include "run_id: 20260516-143946Z"
+End
+
+It "omits empty top-level fields"
+When call emit_yaml \
+  "skill-x" "2026-05-16T00:00:00Z" "deadbee" "false" "" \
+  "" "" "main" "abc1234" "" "" \
+  yaml_keys yaml_values yaml_kinds
+The output should not include "ticket_id"
+The output should not include "ticket_ref"
+The output should not include "pr:"
+The output should not include "run_id"
+End
+
+It "emits scalar extensions after canonical fields"
+yaml_keys+=("title")
+yaml_values[title]="My change"
+yaml_kinds[title]="scalar"
+When call emit_yaml \
+  "summarize-change" "2026-05-16T00:00:00Z" "deadbee" "true" "" \
+  "" "" "main" "abc1234" "" "" \
+  yaml_keys yaml_values yaml_kinds
+The output should include "title: My change"
+End
+
+It "emits flow-list extensions after canonical fields"
+yaml_keys+=("commits")
+yaml_values[commits]="a1b2c3d,e4f5g6h"
+yaml_kinds[commits]="list"
+When call emit_yaml \
+  "create-devlog" "2026-05-16T00:00:00Z" "deadbee" "true" "" \
+  "" "" "main" "abc1234" "" "" \
+  yaml_keys yaml_values yaml_kinds
+The output should include "commits: [a1b2c3d, e4f5g6h]"
+End
+
+emit_three_ordered() {
+  emit_yaml \
+    "summarize-change" "2026-05-16T00:00:00Z" "deadbee" "true" "" \
+    "" "" "main" "abc1234" "" "" \
+    yaml_keys yaml_values yaml_kinds
+}
+
+It "preserves insertion order across mixed scalar and list extensions"
+yaml_keys+=("scope")
+yaml_values[scope]="agents"
+yaml_kinds[scope]="scalar"
+yaml_keys+=("commits")
+yaml_values[commits]="a1b2c3d"
+yaml_kinds[commits]="list"
+yaml_keys+=("type")
+yaml_values[type]="feat"
+yaml_kinds[type]="scalar"
+When call emit_three_ordered
+The output should include "scope: agents"
+The output should include "commits: "
+The output should include "type: feat"
+# Ensure scope < commits < type ordering. Glob brackets need escaping;
+# match the contiguous block including newlines via a structural pattern.
+The output should match pattern "*scope: agents*commits:*type: feat*"
+End
+
+It "auto-quotes values containing colons"
+yaml_keys+=("title")
+yaml_values[title]="Add: feature"
+yaml_kinds[title]="scalar"
+When call emit_yaml \
+  "summarize-change" "2026-05-16T00:00:00Z" "deadbee" "true" "" \
+  "" "" "main" "abc1234" "" "" \
+  yaml_keys yaml_values yaml_kinds
+The output should include "title: 'Add: feature'"
+End
+
+It "omits scalar extensions whose value is empty"
+yaml_keys+=("empty_field")
+yaml_values[empty_field]=""
+yaml_kinds[empty_field]="scalar"
+When call emit_yaml \
+  "summarize-change" "2026-05-16T00:00:00Z" "deadbee" "true" "" \
+  "" "" "main" "abc1234" "" "" \
+  yaml_keys yaml_values yaml_kinds
+The output should not include "empty_field"
+End
+End
+
+Describe "add_override"
+add_override_setup() {
+  unset overrides
+  declare -gA overrides=()
+}
+
+BeforeEach "add_override_setup"
+
+It "records the override when the argument is well-formed"
+add_override "branch=custom" overrides
+When call echo "${overrides[branch]}"
+The output should equal "custom"
+End
+
+It "records an empty value (force-omit) when the argument is KEY="
+add_override "run_id=" overrides
+When call test "${overrides[run_id]+set}" = "set"
+The status should be success
+End
+
+It "fails when the argument has no equals sign"
+When run add_override "bad_arg" overrides
+The status should be failure
+The stderr should include "missing '='"
+End
+
+It "fails when the key is empty"
+When run add_override "=value" overrides
+The status should be failure
+The stderr should include "empty key"
+End
+End
+
+Describe "main"
+setup_main_validation() {
+  tmpdir=$(mktemp -d)
+  pushd "$tmpdir" >/dev/null
+}
+
+cleanup_main_validation() {
+  popd >/dev/null
+  rm -rf "$tmpdir"
+}
+
+BeforeEach "setup_main_validation"
+AfterEach "cleanup_main_validation"
+
+It "exits non-zero with a diagnostic when --skill is missing in yaml mode"
+When run main --format yaml --interactive true
+The status should be failure
+The stderr should include "--skill is required"
+End
+
+It "exits non-zero with a diagnostic when --interactive is missing in yaml mode"
+When run main --format yaml --skill foo
+The status should be failure
+The stderr should include "--interactive is required"
+End
+
+It "exits non-zero with a diagnostic when --format is xml"
+When run main --format xml --skill foo --interactive true
+The status should be failure
+The stderr should include "unknown --format"
+End
+
+It "exits non-zero with a diagnostic when --interactive value is neither true nor false"
+When run main --format yaml --skill foo --interactive maybe
+The status should be failure
+The stderr should include "--interactive must be true or false"
+End
+End
+
+Describe "main end-to-end"
+setup_main_e2e() {
+  tmpdir=$(mktemp -d)
+  pushd "$tmpdir" >/dev/null
+  # Initialize a minimal git repository so `current_branch` and
+  # `git rev-parse --short HEAD` succeed.
+  git init --quiet --initial-branch=main .
+  git config user.email "test@example.com"
+  git config user.name "Test"
+  git commit --allow-empty --quiet -m "initial"
+  # Write the branch manifest the script reads for session-level fields.
+  mkdir -p .agents
+  cat >.agents/main.branch-manifest.json <<'JSON'
+{
+  "platform": "github",
+  "ticket_id": "537",
+  "ticket_ref": "#537",
+  "default_branch": "HEAD"
+}
+JSON
+  # Stub `gh` to return empty (no PR), so the test is hermetic.
+  mkdir -p bin
+  cat >bin/gh <<'SH'
+#!/usr/bin/env bash
+exit 0
+SH
+  chmod +x bin/gh
+  ORIGINAL_PATH="$PATH"
+  PATH="$tmpdir/bin:$PATH"
+}
+
+cleanup_main_e2e() {
+  PATH="$ORIGINAL_PATH"
+  popd >/dev/null
+  rm -rf "$tmpdir"
+}
+
+BeforeEach "setup_main_e2e"
+AfterEach "cleanup_main_e2e"
+
+It "emits extension fields end-to-end and force-omits run_id via --override KEY="
+When run main \
+  --skill foo \
+  --interactive true \
+  --extra "alpha=1" \
+  --extra-list "tags=a,b" \
+  --override "run_id="
+The status should be success
+The output should include "skill: foo"
+The output should include "isInteractive: true"
+The output should include "alpha: 1"
+The output should include "tags: [a, b]"
+The output should not include "run_id"
+End
+End
diff --git a/packages/agents/content/scripts/resolve-frontmatter.sh b/packages/agents/content/scripts/resolve-frontmatter.sh
index d3eba6ec..cec62009 100755
--- a/packages/agents/content/scripts/resolve-frontmatter.sh
+++ b/packages/agents/content/scripts/resolve-frontmatter.sh
@@ -1,43 +1,52 @@
 #!/usr/bin/env bash
-# Emit the canonical artifact-frontmatter fields as a single JSON object.
+# Emit canonical artifact-frontmatter fields as YAML (default) or JSON.
 #
-# Reads `.agents/{sanitized-branch}.branch-manifest.json` (produced by the
-# `get-session-context` skill) for session-level fields and runs git +
-# platform-specific PR lookup for the rest. Skills consume the JSON output
-# to populate the universal portion of their artifact frontmatter without
-# repeating the underlying shell logic.
+# Reads `.agents/{sanitized-branch}.branch-manifest.json` (produced by the `get-session-context` skill) for
+# session-level fields and runs git + platform-specific PR lookup for the rest. Skills consume the output to
+# populate the universal portion of their artifact frontmatter without repeating the underlying shell logic.
 #
 # Usage:
-#   resolve-frontmatter.sh
+#   resolve-frontmatter.sh --skill NAME --interactive true|false [...]
+#   resolve-frontmatter.sh --format json
 #   resolve-frontmatter.sh --help
 #
-# Output (stdout): a JSON object with these keys, all optional except
-# `branch`, `commit`, `platform`, and `timestamp`:
+# Flags:
+#   --skill NAME              provenance.skill value (required in yaml mode).
+#   --interactive true|false  provenance.isInteractive value (required in yaml mode).
+#   --model ID                provenance.model value (optional).
+#   --extra KEY=VALUE         Append a scalar extension key (repeatable).
+#                             Values may contain `=`; split on the first `=`.
+#   --extra-list KEY=v1,v2,…  Append a flow-list extension key (repeatable).
+#                             Values are split on `,`.
+#   --override KEY=VALUE      Force a canonical/provenance field to VALUE.
+#                             Empty VALUE force-omits the key.
+#   --format yaml|json        Output format (default `yaml`).
+#   --help, -h                Show usage and exit 0.
 #
-#   branch        Raw branch name from session context.
-#   commit        Short SHA of HEAD.
-#   baseSha       Short SHA of `origin/{default_branch}`. Omitted if the
-#                 ref is unresolvable (shallow clone, no remote, etc.).
-#   pr            Full PR URL for the current branch. Omitted when no PR
-#                 exists (lookup returned empty) AND on lookup failure
-#                 (timeout, non-zero exit, auth error, etc.).
-#   ticket_id     From session context. Omitted when null.
-#   ticket_ref    From session context. Omitted when null.
-#   platform      `github` or `bitbucket`, from session context.
-#   timestamp     Current UTC ISO 8601 timestamp.
-#   run_id        Active orchestrated-run ID, read from
-#                 `.claude/tmp/active-run-dir` (basename) when present.
-#                 Omitted otherwise.
+# Output (yaml mode, stdout): a complete YAML frontmatter block including
+# `---` delimiters. Field order:
+#
+#   provenance:
+#     skill, timestamp, baseSha, isInteractive, model     (camelCase)
+#   ticket_id, ticket_ref, branch, commit, pr, run_id     (snake_case)
+#   {--extra / --extra-list extensions in insertion order}
+#
+# Output (json mode, stdout): a single JSON object — backward-compatible
+# with prior `--format json` (default) callers. Keys: branch, commit,
+# baseSha, pr, ticket_id, ticket_ref, platform, timestamp, run_id.
+#
+# Omission rules (both formats): any key whose resolved value is empty is
+# omitted entirely. `--override KEY=` with empty value force-omits even
+# when the script resolved a value.
 #
 # Warnings (stderr): when PR resolution fails (not when it returns empty),
 # emits the canonical warning `Note: PR lookup failed; proceeding without
-# pr field.` so the caller can surface it in agent text output. Empty
-# output is the normal no-PR case and emits no warning.
+# pr field.` so the caller can surface it in agent text output.
 #
 # Exit codes:
-#   0  Always when a JSON object is emitted (even with reduced fields).
-#   1  Missing branch manifest — caller must invoke `get-session-context`
-#      first. Not in a git repo. Missing `jq`.
+#   0  Success.
+#   1  Missing branch manifest (run `get-session-context` first), not in a
+#      git repo, missing `jq`, or required-arg violation in yaml mode.
 
 set -euo pipefail
 
@@ -49,26 +58,93 @@ show_usage() {
   local exit_code="${1:-1}"
   cat <<EOF
 Usage:
-  $PROG
+  $PROG --skill NAME --interactive true|false [--model ID]
+       [--extra KEY=VALUE ...] [--extra-list KEY=v1,v2,... ...]
+       [--override KEY=VALUE ...] [--format yaml|json]
+  $PROG --format json
   $PROG --help
 
-Emit canonical artifact-frontmatter fields as a JSON object on stdout.
-See script header for full field list and failure semantics.
+Emit canonical artifact-frontmatter fields as YAML (default) or JSON.
+See script header for field list, ordering, and omission semantics.
 EOF
   exit "$exit_code"
 }
 
 # Main flow
 main() {
-  if [[ "${1:-}" == "--help" || "${1:-}" == "-h" ]]; then
-    show_usage 0
-  fi
+  # -- Argument parsing --
+  local format="yaml"
+  local skill="" model=""
+  local interactive=""
+  local interactive_set=false
+  declare -a extra_keys=()
+  declare -A extra_values=()
+  declare -A extra_kinds=()
+  declare -A overrides=()
+
+  while [[ "$#" -gt 0 ]]; do
+    case "$1" in
+    --help | -h)
+      show_usage 0
+      ;;
+    --format)
+      [[ "$#" -ge 2 ]] || fail "missing value for --format"
+      format="$2"
+      shift 2
+      ;;
+    --skill)
+      [[ "$#" -ge 2 ]] || fail "missing value for --skill"
+      skill="$2"
+      shift 2
+      ;;
+    --interactive)
+      [[ "$#" -ge 2 ]] || fail "missing value for --interactive"
+      interactive="$2"
+      interactive_set=true
+      shift 2
+      ;;
+    --model)
+      [[ "$#" -ge 2 ]] || fail "missing value for --model"
+      model="$2"
+      shift 2
+      ;;
+    --extra)
+      [[ "$#" -ge 2 ]] || fail "missing value for --extra"
+      add_extra "scalar" "$2" extra_keys extra_values extra_kinds
+      shift 2
+      ;;
+    --extra-list)
+      [[ "$#" -ge 2 ]] || fail "missing value for --extra-list"
+      add_extra "list" "$2" extra_keys extra_values extra_kinds
+      shift 2
+      ;;
+    --override)
+      [[ "$#" -ge 2 ]] || fail "missing value for --override"
+      add_override "$2" overrides
+      shift 2
+      ;;
+    *)
+      echo "$PROG: unexpected argument: $1" >&2
+      show_usage 1
+      ;;
+    esac
+  done
+
+  case "$format" in
+  yaml | json) ;;
+  *) fail "unknown --format: $format (expected yaml|json)" ;;
+  esac
 
-  if [[ "$#" -gt 0 ]]; then
-    echo "$PROG: unexpected argument: $1" >&2
-    show_usage 1
+  if [[ "$format" == "yaml" ]]; then
+    [[ -n "$skill" ]] || fail "--skill is required in yaml mode"
+    [[ "$interactive_set" == "true" ]] || fail "--interactive is required in yaml mode"
+    case "$interactive" in
+    true | false) ;;
+    *) fail "--interactive must be true or false (got: $interactive)" ;;
+    esac
   fi
 
+  # -- Resolve session-level values --
   require_commands jq git
 
   local branch
@@ -92,20 +168,90 @@ main() {
   run_id=$(resolve_run_id)
   timestamp=$(date -u +%Y-%m-%dT%H:%M:%SZ)
 
-  emit_json "$branch" "$commit" "$base_sha" "$pr_url" \
-    "$ticket_id" "$ticket_ref" "$platform" "$timestamp" "$run_id"
+  # -- Apply overrides --
+  branch=$(apply_override "branch" "$branch" overrides)
+  commit=$(apply_override "commit" "$commit" overrides)
+  base_sha=$(apply_override "baseSha" "$base_sha" overrides)
+  pr_url=$(apply_override "pr" "$pr_url" overrides)
+  ticket_id=$(apply_override "ticket_id" "$ticket_id" overrides)
+  ticket_ref=$(apply_override "ticket_ref" "$ticket_ref" overrides)
+  platform=$(apply_override "platform" "$platform" overrides)
+  timestamp=$(apply_override "timestamp" "$timestamp" overrides)
+  run_id=$(apply_override "run_id" "$run_id" overrides)
+
+  if [[ "$format" == "json" ]]; then
+    emit_json "$branch" "$commit" "$base_sha" "$pr_url" \
+      "$ticket_id" "$ticket_ref" "$platform" "$timestamp" "$run_id"
+  else
+    emit_yaml \
+      "$skill" "$timestamp" "$base_sha" "$interactive" "$model" \
+      "$ticket_id" "$ticket_ref" "$branch" "$commit" "$pr_url" "$run_id" \
+      extra_keys extra_values extra_kinds
+  fi
+}
+
+# Parses a `KEY=VALUE` argument into the caller's `key` and `value` variables.
+# The `flag` argument is used only for error messages. Fails when `=` is missing or the key portion is empty.
+parse_key_value() {
+  local arg="$1" flag="$2"
+  local -n key_out="$3"
+  local -n value_out="$4"
+  if [[ "$arg" != *"="* ]]; then
+    fail "$flag argument missing '=': $arg"
+  fi
+  key_out="${arg%%=*}"
+  value_out="${arg#*=}"
+  [[ -n "$key_out" ]] || fail "$flag argument has empty key"
 }
 
-# Print short SHA of `default_branch` (e.g., `origin/main`) or empty if
-# unresolvable. A shallow clone or missing remote silently degrades to empty.
+# Appends an extension key/value to the caller's ordered list. Splits the
+# argument once on the first `=`. `kind` is either `scalar` or `list`.
+add_extra() {
+  local kind="$1" arg="$2"
+  local -n keys_ref="$3"
+  local -n values_ref="$4"
+  local -n kinds_ref="$5"
+  local key value
+  parse_key_value "$arg" "--extra/--extra-list" key value
+  if [[ -z "${kinds_ref[$key]:-}" ]]; then
+    keys_ref+=("$key")
+  else
+    warn "duplicate --extra/--extra-list key '$key': replacing previous value"
+  fi
+  values_ref["$key"]="$value"
+  kinds_ref["$key"]="$kind"
+}
+
+# Records an override key=value. Empty value force-omits the key on emit.
+add_override() {
+  local arg="$1"
+  local -n overrides_ref="$2"
+  local key value
+  parse_key_value "$arg" "--override" key value
+  overrides_ref["$key"]="$value"
+}
+
+# Returns the overridden value when the key has been overridden, otherwise the resolved value.
+# The empty-string override force-omits.
+apply_override() {
+  local key="$1" resolved="$2"
+  local -n overrides_ref="$3"
+  if [[ -n "${overrides_ref[$key]+set}" ]]; then
+    printf '%s' "${overrides_ref[$key]}"
+  else
+    printf '%s' "$resolved"
+  fi
+}
+
+# Prints short SHA of `default_branch` (e.g., `origin/main`) or empty if unresolvable.
+# A shallow clone or missing remote silently degrades to empty.
 resolve_base_sha() {
   local ref="$1"
   git rev-parse --short "$ref" 2>/dev/null || true
 }
 
-# Print the PR URL for the current branch, or empty if no PR or lookup
-# failed. Distinguishes empty-output (silent) from failure (canonical
-# warning emitted to stderr).
+# Prints the PR URL for the current branch, or empty if no PR or lookup failed.
+# Distinguishes empty-output (silent) from failure (canonical warning emitted to stderr).
 resolve_pr_url() {
   local platform="$1"
   local branch="$2"
@@ -116,8 +262,8 @@ resolve_pr_url() {
   esac
 }
 
-# GitHub PR lookup via `gh pr list`. Uses --state all so closed and merged
-# PRs are resolvable for post-merge artifact writes.
+# Looks up a GitHub PR via `gh pr list`.
+# Uses --state all so closed and merged PRs are resolvable for post-merge artifact writes.
 resolve_github_pr() {
   local branch="$1"
   if ! command -v gh >/dev/null 2>&1; then
@@ -135,7 +281,7 @@ resolve_github_pr() {
   printf '%s' "$result"
 }
 
-# Bitbucket PR lookup via Bitbucket Cloud REST API.
+# Looks up a Bitbucket PR via Bitbucket Cloud REST API.
 resolve_bitbucket_pr() {
   local branch="$1"
   if ! command -v curl >/dev/null 2>&1; then
@@ -165,9 +311,8 @@ resolve_bitbucket_pr() {
   printf '%s' "$result"
 }
 
-# Resolve a Bitbucket auth header from environment or macOS keychain. Echoes
-# the full `Authorization:` header value or returns non-zero when no
-# credentials are available.
+# Resolves a Bitbucket auth header from environment or macOS keychain.
+# Echoes the full `Authorization:` header value or returns non-zero when no credentials are available.
 bitbucket_auth_header() {
   if [[ -n "${BITBUCKET_BOT_USERNAME:-}" && -n "${BITBUCKET_BOT_TOKEN:-}" ]]; then
     local basic
@@ -189,18 +334,16 @@ bitbucket_auth_header() {
   return 1
 }
 
-# Emit the canonical PR-lookup-failed warning to stderr. The first argument
-# is an internal diagnostic that is also written to stderr after the
-# canonical line so debug context is available without affecting callers
-# that match the canonical phrasing.
+# Emits the canonical PR-lookup-failed warning to stderr.
+# The first argument is an internal diagnostic that is also written to stderr after the canonical line,
+# so that debug context is available without affecting callers that match the canonical phrasing.
 warn_pr_failure() {
   echo "$CANONICAL_WARNING" >&2
   echo "  ($1)" >&2
 }
 
-# Resolve the active run ID by reading the breadcrumb written by the
-# orchestrate engine at `.claude/tmp/active-run-dir`. Empty when no
-# orchestrated run is active.
+# Resolves the active run ID by reading the breadcrumb written by the orchestrate engine at
+# `.claude/tmp/active-run-dir`. Empty when no orchestrated run is active.
 resolve_run_id() {
   local breadcrumb=".claude/tmp/active-run-dir"
   [[ -r "$breadcrumb" ]] || return 0
@@ -209,8 +352,8 @@ resolve_run_id() {
   basename "$run_dir"
 }
 
-# Run a command with a timeout. Uses `timeout` or `gtimeout` when available,
-# otherwise runs without a timeout (the agent's Bash-tool timeout still
+# Run a command with a timeout.
+# Uses `timeout` or `gtimeout` when available; otherwise, runs without a timeout (the agent's Bash-tool timeout still
 # applies). The command and its arguments are passed verbatim.
 run_with_timeout() {
   local secs="$1"
@@ -224,12 +367,12 @@ run_with_timeout() {
   fi
 }
 
-# Get the current branch name. Returns non-zero outside a git repository.
+# Gets the current branch name. Returns non-zero outside a git repository.
 current_branch() {
   git rev-parse --abbrev-ref HEAD 2>/dev/null
 }
 
-# Read the branch manifest for the given branch. Echoes the JSON content.
+# Reads the branch manifest for the given branch. Echoes the JSON content.
 # Returns non-zero when the manifest is missing.
 read_manifest() {
   local branch="$1"
@@ -240,8 +383,8 @@ read_manifest() {
   cat "$path"
 }
 
-# Sanitize a branch name for filesystem use: replace `/` with `-` and trim
-# any trailing `-` characters. Mirrors `get-session-context` behavior.
+# Sanitizes a branch name for filesystem use: replace `/` with `-` and trims any trailing `-` characters.
+# Mirrors `get-session-context` behavior.
 sanitize_branch() {
   local branch="$1"
   branch="${branch//\//-}"
@@ -249,9 +392,8 @@ sanitize_branch() {
   printf '%s' "$branch"
 }
 
-# Construct the JSON output from resolved values. Optional fields are
-# omitted (rather than emitted as null or empty) so consumers can rely on
-# `has(field)` semantics.
+# Constructs the JSON output from resolved values.
+# Optional fields are omitted (rather than emitted as null or empty) so consumers can rely on `has(field)` semantics.
 emit_json() {
   local branch="$1" commit="$2" base_sha="$3" pr_url="$4"
   local ticket_id="$5" ticket_ref="$6" platform="$7" timestamp="$8" run_id="$9"
@@ -281,8 +423,142 @@ emit_json() {
     '
 }
 
-# Require each named command to be on PATH. Exits 1 with a clear message
-# when one is missing.
+# Emits the canonical YAML frontmatter block, including `---` delimiters. Empty values are omitted.
+# Field order is fixed: Provenance block, canonical top-level fields, caller-supplied extensions in insertion order.
+emit_yaml() {
+  local skill="$1" timestamp="$2" base_sha="$3" interactive="$4" model="$5"
+  local ticket_id="$6" ticket_ref="$7" branch="$8" commit="$9" pr_url="${10}" run_id="${11}"
+  local -n yaml_extra_keys="${12}"
+  local -n yaml_extra_values="${13}"
+  local -n yaml_extra_kinds="${14}"
+
+  printf '%s\n' "---"
+
+  # Provenance block
+  printf '%s\n' "provenance:"
+  emit_yaml_indented_scalar "skill" "$skill"
+  emit_yaml_indented_scalar "timestamp" "$timestamp"
+  [[ -n "$base_sha" ]] && emit_yaml_indented_scalar "baseSha" "$base_sha"
+  # isInteractive is a boolean — emit unquoted true|false.
+  printf '  %s: %s\n' "isInteractive" "$interactive"
+  [[ -n "$model" ]] && emit_yaml_indented_scalar "model" "$model"
+
+  # Canonical top-level fields
+  [[ -n "$ticket_id" ]] && emit_yaml_scalar "ticket_id" "$ticket_id"
+  [[ -n "$ticket_ref" ]] && emit_yaml_scalar "ticket_ref" "$ticket_ref"
+  emit_yaml_scalar "branch" "$branch"
+  emit_yaml_scalar "commit" "$commit"
+  [[ -n "$pr_url" ]] && emit_yaml_scalar "pr" "$pr_url"
+  [[ -n "$run_id" ]] && emit_yaml_scalar "run_id" "$run_id"
+
+  # Extension fields in insertion order
+  local key kind value
+  for key in "${yaml_extra_keys[@]+"${yaml_extra_keys[@]}"}"; do
+    value="${yaml_extra_values[$key]}"
+    kind="${yaml_extra_kinds[$key]}"
+    if [[ "$kind" == "list" ]]; then
+      # Flow-list extensions intentionally emit `key: []` for empty values
+      # (see `emit_yaml_flow_list`); only scalar extensions are subject to the canonical-field omission rule.
+      emit_yaml_flow_list "$key" "$value"
+    else
+      [[ -n "$value" ]] || continue
+      emit_yaml_scalar "$key" "$value"
+    fi
+  done
+
+  printf '%s\n' "---"
+}
+
+# Emit a top-level YAML `key: value` line, auto-quoting the value as needed.
+emit_yaml_scalar() {
+  local key="$1" value="$2"
+  printf '%s: %s\n' "$key" "$(yaml_quote "$value")"
+}
+
+# Emit an indented YAML `key: value` line inside the provenance block.
+emit_yaml_indented_scalar() {
+  local key="$1" value="$2"
+  printf '  %s: %s\n' "$key" "$(yaml_quote "$value")"
+}
+
+# Emits a top-level YAML flow list: `key: [v1, v2, v3]`.
+# Empty value emits an empty flow list `key: []`. Elements are split on `,` and each is passed through `yaml_quote`.
+emit_yaml_flow_list() {
+  local key="$1" raw="$2"
+  if [[ -z "$raw" ]]; then
+    printf '%s: []\n' "$key"
+    return
+  fi
+  local IFS=','
+  # Disable globbing around the word-split so that list elements containing glob metacharacters (`*`, `?`, `[...]`)
+  # are not expanded against the filesystem before `yaml_quote` sees them.
+  set -f
+  # shellcheck disable=SC2206
+  local -a parts=( $raw )
+  set +f
+  local out="" i
+  for ((i = 0; i < ${#parts[@]}; i++)); do
+    if [[ "$i" -gt 0 ]]; then
+      out+=", "
+    fi
+    out+="$(yaml_quote "${parts[$i]}")"
+  done
+  printf '%s: [%s]\n' "$key" "$out"
+}
+
+# Returns the value either bare or single-quoted depending on YAML auto-quoting rules.
+# The predicate quotes when the value
+# - contains any of the unsafe glyphs (`# : [ ] { } , & * ! | > < ? % @ \` ` `),
+# - has leading or trailing whitespace,
+# - is empty, or
+# - begins with `-` / `?` / `:`.
+# Embedded single quotes are doubled inside the quoted form.
+yaml_quote() {
+  local v="$1"
+  if needs_yaml_quoting "$v"; then
+    local escaped="${v//\'/\'\'}"
+    printf "'%s'" "$escaped"
+  else
+    printf '%s' "$v"
+  fi
+}
+
+# Decides whether `v` needs single-quote wrapping. Returns 0 (true) when quoting is required, 1 otherwise.
+#
+# YAML parses `:` as a key indicator only when followed by whitespace or end-of-value,
+# so URLs like `https://...` are safe bare. `#` is always treated as a comment introducer in YAML 1.1/1.2 (the spec is
+# permissive about the preceding context), so we quote any value containing `#`.
+needs_yaml_quoting() {
+  local v="$1"
+  # Empty values must be quoted.
+  [[ -z "$v" ]] && return 0
+  # Leading or trailing whitespace.
+  [[ "$v" =~ ^[[:space:]] ]] && return 0
+  [[ "$v" =~ [[:space:]]$ ]] && return 0
+  # Leading sigils that YAML interprets specially. `?` as a leading
+  # character is handled by the glyphs-anywhere case below.
+  case "$v" in
+  -* | :*) return 0 ;;
+  esac
+  # Colon followed by whitespace anywhere is a key indicator.
+  [[ "$v" =~ :[[:space:]] ]] && return 0
+  # Trailing colon is the value-end form of a key indicator.
+  [[ "$v" == *: ]] && return 0
+  # Glyphs anywhere in the value that demand quoting.
+  case "$v" in
+  *'#'* | *'['* | *']'* | *'{'* | *'}'* \
+    | *','* | *'&'* | *'*'* | *'!'* | *'|'* | *'>'* | *'<'* \
+    | *'?'* | *'%'* | *'@'* | *'`'* | *"'"* | *'"'*)
+    return 0
+    ;;
+  esac
+  # Values that look like YAML booleans/null or numbers stay bare per the current predicate.
+  # The predicate's value space is the schema's canonical fields and known extensions;
+  # expand here if a future extension introduces ambiguity.
+  return 1
+}
+
+# Require each named command to be on PATH. Exits 1 with a clear message when one is missing.
 require_commands() {
   local cmd
   for cmd in "$@"; do
@@ -296,6 +572,13 @@ fail() {
   exit 1
 }
 
+# Prints a non-fatal warning to stderr.
+# Used for soft-failure conditions where the script should continue with degraded behavior (e.g., a duplicate
+# optional-extension key that overwrites an earlier value).
+warn() {
+  echo "$PROG: warning: $1" >&2
+}
+
 # Guard against execution when sourced (e.g., from shellspec tests).
 if [[ "${BASH_SOURCE[0]}" == "${0}" ]]; then
   main "$@"
diff --git a/packages/agents/content/skills/_data/artifact-conventions.md b/packages/agents/content/skills/_data/artifact-conventions.md
index bf89f028..1bc402de 100644
--- a/packages/agents/content/skills/_data/artifact-conventions.md
+++ b/packages/agents/content/skills/_data/artifact-conventions.md
@@ -205,6 +205,15 @@ The table below lists only the universal fields. Artifact-specific extensions (`
 
 Skills resolve `pr` at write time via the shared dispatch documented in [`pr-resolution.md`](pr-resolution.md). On failure, the `pr:` line is omitted and the skill emits the canonical warning text — the artifact write itself is never blocked.
 
+### Bespoke frontmatter composition
+
+Most skills and subagents produce frontmatter by running `resolve-frontmatter.sh` in its default YAML mode and prepending the output verbatim. Two sites are deliberate exceptions and opt into `--format json` to compose the YAML block themselves:
+
+- `refine-plan` — the `provenance:` block is case-branched on the input artifact's existing provenance (preserving `skill`, `baseSha`, `isInteractive`, and `iteration` from the original authoring skill, with fallbacks when the input has no provenance). The shell flag surface cannot express this conditional logic cleanly.
+- `wrap-up` (deferred-findings artifact) — `tickets_created` is a list of `{id, items}` objects, a structure that has no clean CLI expression and is best composed in the skill's own logic.
+
+These two sites read the script's JSON output, then write the YAML frontmatter themselves. The pattern is intentional, not a workaround — keep new skills on the YAML mode path unless they have a similarly structural reason to deviate.
+
 ## Plan provenance
 
 This artifact uses the [universal artifact frontmatter](#universal-artifact-frontmatter) plus the following artifact-specific extension:
diff --git a/packages/agents/content/skills/_data/pr-resolution.md b/packages/agents/content/skills/_data/pr-resolution.md
index 3bf504d1..2ee5aa70 100644
--- a/packages/agents/content/skills/_data/pr-resolution.md
+++ b/packages/agents/content/skills/_data/pr-resolution.md
@@ -1,13 +1,13 @@
 # PR resolution
 
-Shared contract for resolving the `pr:` frontmatter field at artifact write time. Skills do not invoke `gh` or `curl` directly — they call `scripts/resolve-frontmatter.sh`, which implements this contract once and emits the resolved `pr:` value (along with all other canonical frontmatter fields) as part of its JSON output. This document defines the contract the script obeys; consumers do not embed dispatch snippets directly.
+Shared contract for resolving the `pr:` frontmatter field at artifact write time. Skills do not invoke `gh` or `curl` directly — they call `scripts/resolve-frontmatter.sh`, which implements this contract once and emits the resolved `pr:` value (along with all other canonical frontmatter fields) as part of its output. This document defines the contract the script obeys; consumers do not embed dispatch snippets directly.
 
 ## Contract
 
 - Resolution runs at artifact write time, against the current branch.
 - The Bash invocation uses a **5-second timeout** (`timeout: 5000` on the Bash tool when calling the script; the script itself wraps the platform CLI in `timeout`/`gtimeout` when available).
-- On **empty output** (no PR exists for the branch — the lookup succeeded), the script omits the `pr` field from its JSON output. Consumers must omit the `pr:` line from frontmatter. Do **not** emit a warning. This is the normal pre-PR case.
-- On **failure** (CLI unavailable, auth error, network error, timeout, or any non-zero exit), the script omits the `pr` field from its JSON output and emits the canonical warning to stderr. Consumers must omit the `pr:` line **and** surface the warning in the agent's text output. **Never block the artifact write.**
+- On **empty output** (no PR exists for the branch — the lookup succeeded), the script omits the `pr` field from its output. Consumers must omit the `pr:` line from frontmatter. Do **not** emit a warning. This is the normal pre-PR case.
+- On **failure** (CLI unavailable, auth error, network error, timeout, or any non-zero exit), the script omits the `pr` field from its output and emits the canonical warning to stderr. Consumers must omit the `pr:` line **and** surface the warning in the agent's text output. **Never block the artifact write.**
 
 ### Canonical warning text
 
diff --git a/packages/agents/content/skills/create-devlog/SKILL.md b/packages/agents/content/skills/create-devlog/SKILL.md
index 46fe2030..bd865aa8 100644
--- a/packages/agents/content/skills/create-devlog/SKILL.md
+++ b/packages/agents/content/skills/create-devlog/SKILL.md
@@ -17,24 +17,9 @@ Summarize changes made in recent commits or the working tree.
 
 ## Output format
 
-The devlog file begins with YAML frontmatter conforming to the canonical schema (see [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) and the devlog-specific section, [Devlog frontmatter](../_data/artifact-conventions.md#devlog-frontmatter)) followed by the markdown body:
+The devlog file begins with YAML frontmatter conforming to the canonical schema; see the canonical example in [artifact-conventions.md](../_data/artifact-conventions.md#universal-artifact-frontmatter) and the devlog-specific extensions in [Devlog frontmatter](../_data/artifact-conventions.md#devlog-frontmatter). The body following the frontmatter has this structure:
 
 ```markdown
----
-provenance:
-  skill: create-devlog
-  timestamp: 2026-04-18T15:30:00Z
-  baseSha: 4f8b158
-  isInteractive: true
-ticket_id: '426'
-ticket_ref: '#426'
-run_id: 20260419-012539Z
-branch: 426/feat/example-branch
-commit: 1d2c3b4
-pr: https://github.com/williamthorsen/codeassembly/pull/591
-commits: [a1b2c3d, e4f5g6h]
----
-
 # Devlog: {Concise description}
 
 **Date**: {YYYY-MM-DD HH:MM UTC}
@@ -97,16 +82,28 @@ Follow [artifact conventions](../_data/artifact-conventions.md).
 
 The devlog frontmatter conforms to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema plus the devlog-specific extensions listed in [Devlog frontmatter](../_data/artifact-conventions.md#devlog-frontmatter).
 
-<!-- include: ../../_partials/frontmatter-via-script.md -->
+Resolve `$run_id_arg` from the `--run-id={id}` argument (empty when not supplied). Resolve the `commits` extension according to the mode:
+
+- No argument (last commit): `commits_arg=$(git log -n 1 --format=%h)`.
+- `<n>` (last N commits): `commits_arg=$(git log -n N --format=%h | paste -sd, -)`.
+- `working-tree`: Do not pass `--extra-list commits=...`.
+
+Run via Bash, substituting the resolved arguments:
+
+```bash
+resolve-frontmatter.sh \
+  --skill create-devlog \
+  --interactive true \
+  --model "$MODEL_ID" \
+  ${commits_arg:+--extra-list "commits=$commits_arg"} \
+  --override "run_id=$run_id_arg"
+```
+
+The `${commits_arg:+--extra-list "commits=$commits_arg"}` form expands to the flag only when `$commits_arg` is non-empty, so the `working-tree` mode (where `$commits_arg` is unset) naturally omits the `commits` field rather than emitting `commits: []`. Quoting `run_id=$run_id_arg` ensures the empty-value force-omit case works when no `--run-id` was supplied.
+
+Prepend the script's output verbatim to the artifact body. Source `$MODEL_ID` from your system-prompt environment block: the line `model named ... model ID is ...`.
 
-- `provenance.skill`: always `create-devlog`.
-- `provenance.isInteractive`: always `true`.
-- `run_id`: **override** the script's value. Emit only if `--run-id={id}` was supplied as an argument; the caller is the source of truth. Do not use the script's breadcrumb-derived value, and do not perform any filesystem discovery.
-- `commits` (devlog-specific extension; distinct from the universal `commit` field):
-  - No argument (last commit): single short SHA from `git log -n 1 --format=%h`.
-  - `<n>` (last N commits): list of N short SHAs from `git log -n {N} --format=%h`.
-  - `working-tree`: omit the `commits` field entirely.
-  <!-- /include -->
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ### Filename examples
 
diff --git a/packages/agents/content/skills/create-ticket/SKILL.md b/packages/agents/content/skills/create-ticket/SKILL.md
index 8065f961..2a6250f9 100644
--- a/packages/agents/content/skills/create-ticket/SKILL.md
+++ b/packages/agents/content/skills/create-ticket/SKILL.md
@@ -151,12 +151,11 @@ If a plan is also saved in step 7, it uses the same frontmatter shape with `prov
 
 The artifact's frontmatter conforms to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema.
 
-<!-- include: ../../_partials/frontmatter-via-script.md -->
+Source `$MODEL_ID` from your system-prompt environment block: the line `model named ... model ID is ...`.
 
-- `provenance.skill`: always `create-ticket`.
-- `provenance.isInteractive`: always `true`.
-- `provenance.model`: the model identifier you are executing under. Read this from your system-prompt environment block — the line `model named ... model ID is ...`.
-<!-- /include -->
+Run `resolve-frontmatter.sh --skill create-ticket --interactive true --model "$MODEL_ID"` via Bash. Prepend the output verbatim to the artifact body.
+
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ### 7. Save plan (if present)
 
diff --git a/packages/agents/content/skills/design-and-plan/SKILL.md b/packages/agents/content/skills/design-and-plan/SKILL.md
index c5fea1c5..26add7d9 100644
--- a/packages/agents/content/skills/design-and-plan/SKILL.md
+++ b/packages/agents/content/skills/design-and-plan/SKILL.md
@@ -204,10 +204,9 @@ Present the plan to the user. Revise until approved.
 
 2. Resolve frontmatter fields for both artifacts. The frontmatter conforms to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema.
 
-   <!-- include: ../../_partials/frontmatter-via-script.md -->
-   - `provenance.skill`: always `design-and-plan`.
-   - `provenance.isInteractive`: always `true`.
-   <!-- /include -->
+   Run `resolve-frontmatter.sh --skill design-and-plan --interactive true` via Bash. Prepend the output verbatim to each artifact body.
+
+   If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 3. Save both artifacts following `save-artifact` naming conventions:
    - Ticket: `{YYYYMMDD-HHMMSSZ}_{slug}_ticket.md`
diff --git a/packages/agents/content/skills/ex-post-facto/SKILL.md b/packages/agents/content/skills/ex-post-facto/SKILL.md
index 88d62673..9b4690e8 100644
--- a/packages/agents/content/skills/ex-post-facto/SKILL.md
+++ b/packages/agents/content/skills/ex-post-facto/SKILL.md
@@ -20,23 +20,9 @@ git diff $DEFAULT_BRANCH...HEAD
 
 ## Output structure
 
-The artifact begins with YAML frontmatter conforming to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema. See [Frontmatter resolution](#frontmatter-resolution) below for field resolution.
+The artifact begins with YAML frontmatter conforming to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema. See [Frontmatter resolution](#frontmatter-resolution) below for field resolution. The frontmatter conforms to the canonical schema; see the canonical example in [artifact-conventions.md](../_data/artifact-conventions.md#universal-artifact-frontmatter).
 
 ```markdown
----
-provenance:
-  skill: ex-post-facto
-  timestamp: '{ISO 8601 UTC timestamp}'
-  baseSha: '{short SHA of origin/main, omit if unresolvable}'
-  isInteractive: true
-  model: '{model id}'
-ticket_id: '{ticket id, omit if absent}'
-ticket_ref: '{ticket display ref, omit if absent}'
-branch: '{current branch name}'
-commit: '{short hash of HEAD}'
-pr: '{full PR URL, omit if not resolved}'
----
-
 # {Title}
 
 ## Description
@@ -79,12 +65,11 @@ pr: '{full PR URL, omit if not resolved}'
 
 The artifact's frontmatter conforms to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema.
 
-<!-- include: ../../_partials/frontmatter-via-script.md -->
+Source `$MODEL_ID` from your system-prompt environment block: the line `model named ... model ID is ...`.
+
+Run `resolve-frontmatter.sh --skill ex-post-facto --interactive true --model "$MODEL_ID"` via Bash. Prepend the output verbatim to the artifact body.
 
-- `provenance.skill`: always `ex-post-facto`.
-- `provenance.isInteractive`: always `true`.
-- `provenance.model`: the model identifier you are executing under. Read this from your system-prompt environment block — the line `model named ... model ID is ...`.
-<!-- /include -->
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ## Saving
 
diff --git a/packages/agents/content/skills/orchestrate/SKILL.md b/packages/agents/content/skills/orchestrate/SKILL.md
index dd76ad4c..5b0f5ce4 100644
--- a/packages/agents/content/skills/orchestrate/SKILL.md
+++ b/packages/agents/content/skills/orchestrate/SKILL.md
@@ -254,23 +254,9 @@ Before writing each artifact: format `{seq}` as two zero-padded digits (`{NN}`),
 - **Skipped or conditional artifacts**: do not consume a sequence number. `{seq}` only increments when an artifact is actually written.
 - **Subagents**: receive the full write-target path as an argument. They do not manage sequence numbers themselves.
 
-5. **Write run-manifest artifact** to `{run-dir}/{NN}_orchestrator_run-manifest.md`. The artifact begins with YAML frontmatter conforming to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema (resolved per the [run-manifest and run-summary frontmatter resolution](#run-manifest-and-run-summary-frontmatter-resolution) section below):
+5. **Write run-manifest artifact** to `{run-dir}/{NN}_orchestrator_run-manifest.md`. The artifact begins with YAML frontmatter conforming to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema (resolved per the [run-manifest and run-summary frontmatter resolution](#run-manifest-and-run-summary-frontmatter-resolution) section below). The frontmatter conforms to the canonical schema; see the canonical example in [artifact-conventions.md](../_data/artifact-conventions.md#universal-artifact-frontmatter).
 
 ```markdown
----
-provenance:
-  skill: orchestrate
-  timestamp: '{ISO 8601 UTC timestamp}'
-  baseSha: '{short SHA of origin/main, omit if unresolvable}'
-  isInteractive: false
-ticket_id: '{ticket id, omit if absent}'
-ticket_ref: '{ticket display ref, omit if absent}'
-branch: '{current branch name}'
-commit: '{short hash of HEAD}'
-pr: '{full PR URL, omit if not resolved}'
-run_id: '{run id}'
----
-
 # Run manifest
 
 | Field        | Value              |
@@ -346,9 +332,9 @@ The `{externalPlan}` flag and extracted plan content were already determined in
 
 Emit a `phase_decision` event for every known phase. Iterate through the complete set of known phases (`architecture`, `planning`, `implementation`, `review-cycle`) — not just the phases present in the pipeline:
 
-1. **Phase absent from pipeline**: emit `phase_decision` with `run: false` and `reason: "absent"`.
-2. **Phase present with requirement `required`**: phase always runs. Emit `phase_decision` with `run: true` and `reason: "executed"`.
-3. **Phase present with requirement `optional`**: apply phase-specific skip logic (below). Emit `phase_decision` with `run: true/false` and `reason: "executed"` or `"skipped: {reason}"`.
+1. **Phase absent from pipeline**: Emit `phase_decision` with `run: false` and `reason: "absent"`.
+2. **Phase present with requirement `required`**: Phase always runs. Emit `phase_decision` with `run: true` and `reason: "executed"`.
+3. **Phase present with requirement `optional`**: Apply phase-specific skip logic (below). Emit `phase_decision` with `run: true/false` and `reason: "executed"` or `"skipped: {reason}"`.
 
 For each phase decision, call MCP tool `emit_event`:
 
@@ -388,11 +374,11 @@ When `{planTrust}` is `"high"` and Planning is skipped, the orchestrator produce
 1. **Check for JSON companion:** If the external plan file has a JSON companion (same directory, same base name or `orchestration-plan.json`), read it and use it as `{plan-json-content}`. Skip markdown parsing — the JSON is already structured.
 
 2. **Parse markdown to JSON** (if no companion): Parse the external plan's `### Task N:` sections. For each task section, extract:
-   - `title`: text after `### Task N: `
-   - `files`: lines under `**Files:**` (strip `- Create: `, `- Modify: `, `- Test: ` prefixes)
-   - `dependsOn`: parse `**Depends on:** Step N` or `**Depends on:** Steps N, M` references, converting to integer IDs
-   - `acceptanceCriteria`: bullet items under `**Acceptance criteria:**`
-   - `description`: remaining text in the section (between the title and the first recognized sub-heading)
+   - `title`: Text after `### Task N: `
+   - `files`: Lines under `**Files:**` (strip `- Create: `, `- Modify: `, `- Test: ` prefixes)
+   - `dependsOn`: Parse `**Depends on:** Step N` or `**Depends on:** Steps N, M` references, converting to integer IDs
+   - `acceptanceCriteria`: Bullet items under `**Acceptance criteria:**`
+   - `description`: Remaining text in the section (between the title and the first recognized sub-heading)
 
    If a task section lacks any of these sub-headings, use empty values: `[]` for arrays, `""` for strings.
 
@@ -414,7 +400,7 @@ When `{planTrust}` is `"high"` and Planning is skipped, the orchestrator produce
    }
    ```
 
-3. **Guard: zero-steps fallback.** If `{plan-json-content}` has an empty `steps` array (length 0):
+3. **Guard: Zero-steps fallback.** If `{plan-json-content}` has an empty `steps` array (length 0):
 
    a. Emit a corrective `phase_decision` event for planning:
 
@@ -422,7 +408,7 @@ When `{planTrust}` is `"high"` and Planning is skipped, the orchestrator produce
    Call MCP tool emit_event with:
      runDir: {run-dir}
      event: { event: "phase_decision", phase: "planning", run: true,
-              reason: "fallback: high-trust plan produced zero steps — running planning in adoption mode" }
+              reason: "fallback: High-trust plan produced zero steps — running planning in adoption mode" }
    ```
 
    b. Set `{planTrust}` to `"medium"`. Do not increment `{seq}`.
@@ -458,7 +444,7 @@ When `{planTrust}` is `"high"` and Planning is skipped, the orchestrator produce
    phase: initialization
    ```
 
-6. **Store paths:** Store full paths as `{plan-md-path}` and `{plan-json-path}` for downstream phases. Note: the `phase_decision` for planning was already emitted in the "Skip logic" section above with `reason: "skipped: high-trust plan (skill: {provenance.skill}, baseSha matches main)"`. Do not emit a second `phase_decision` here.
+6. **Store paths:** Store full paths as `{plan-md-path}` and `{plan-json-path}` for downstream phases. Note: The `phase_decision` for planning was already emitted in the "Skip logic" section above with `reason: "skipped: high-trust plan (skill: {provenance.skill}, baseSha matches main)"`. Do not emit a second `phase_decision` here.
 
 ## Authority hierarchy
 
@@ -663,23 +649,9 @@ Dispatch the savings-analyzer subagent as a background Task and immediately proc
     - `ticket_id` and `ticket_ref` — from session context. Omit either when null.
     - `run_id` — the run ID for the current orchestrated run.
 
-Write run-summary artifact to `{run-dir}/{NN}_orchestrator_run-summary.md`. The artifact begins with YAML frontmatter conforming to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema:
+Write run-summary artifact to `{run-dir}/{NN}_orchestrator_run-summary.md`. The artifact begins with YAML frontmatter conforming to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema. The frontmatter conforms to the canonical schema; see the canonical example in [artifact-conventions.md](../_data/artifact-conventions.md#universal-artifact-frontmatter).
 
 ```markdown
----
-provenance:
-  skill: orchestrate
-  timestamp: '{ISO 8601 UTC timestamp}'
-  baseSha: '{short SHA of origin/main, omit if unresolvable}'
-  isInteractive: false
-ticket_id: '{ticket id, omit if absent}'
-ticket_ref: '{ticket display ref, omit if absent}'
-branch: '{current branch name}'
-commit: '{short hash of HEAD}'
-pr: '{full PR URL, omit if not resolved}'
-run_id: '{run id}'
----
-
 # Orchestration summary
 
 ## Task
@@ -743,11 +715,9 @@ Include:
 
 This section governs the frontmatter resolution for both orchestrator-written artifacts — the run-manifest (step 5) and the run-summary (Phase 5) — which use identical field-resolution logic.
 
-<!-- include: ../../_partials/frontmatter-via-script.md -->
+Run `resolve-frontmatter.sh --skill orchestrate --interactive false` via Bash. Prepend the output verbatim to the artifact body.
 
-- `provenance.skill`: always `orchestrate`.
-- `provenance.isInteractive`: always `false`.
-<!-- /include -->
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 The orchestrator's `provenance.model` is omitted — the run-summary aggregates work from many subagents, each with its own model recorded in its own artifact. The summary itself is composed by the orchestrator and is not a single-model artifact.
 
diff --git a/packages/agents/content/skills/plan-orchestrable-steps/SKILL.md b/packages/agents/content/skills/plan-orchestrable-steps/SKILL.md
index 0b5e1f3e..50284f8f 100644
--- a/packages/agents/content/skills/plan-orchestrable-steps/SKILL.md
+++ b/packages/agents/content/skills/plan-orchestrable-steps/SKILL.md
@@ -86,10 +86,9 @@ When the user approves the plan:
 
 1. Resolve frontmatter fields. The frontmatter conforms to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema.
 
-   <!-- include: ../../_partials/frontmatter-via-script.md -->
-   - `provenance.skill`: always `plan-orchestrable-steps`.
-   - `provenance.isInteractive`: always `true`.
-   <!-- /include -->
+   Run `resolve-frontmatter.sh --skill plan-orchestrable-steps --interactive true` via Bash. Prepend the output verbatim to the artifact body.
+
+   If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 2. Add a frontmatter header to the latest plan markdown snapshot. List `{artifact-dir}/*_planner_orchestration-plan.md` files, sort lexicographically descending, and take the first (most recent by timestamp prefix). If no matching files are found, skip the frontmatter header step — the planner did not produce a markdown snapshot. Read the file, prepend the resolved frontmatter, and write back.
 
diff --git a/packages/agents/content/skills/plan/SKILL.md b/packages/agents/content/skills/plan/SKILL.md
index 2e859969..75fe3df6 100644
--- a/packages/agents/content/skills/plan/SKILL.md
+++ b/packages/agents/content/skills/plan/SKILL.md
@@ -15,23 +15,11 @@ Create a structured plan document for analysis, design, or implementation work.
 
 ## Output format
 
-The plan begins with YAML frontmatter conforming to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema and the [plan provenance](../_data/artifact-conventions.md#plan-provenance) extension. `provenance.model` is omitted — plans authored via this skill are co-authored interactively, not solely AI-generated.
+The plan begins with YAML frontmatter conforming to the canonical schema; see the canonical example in [artifact-conventions.md](../_data/artifact-conventions.md#universal-artifact-frontmatter) and the [plan provenance](../_data/artifact-conventions.md#plan-provenance) extension; field-resolution steps live in the [Frontmatter resolution](#frontmatter-resolution) section below. `provenance.model` is omitted — plans authored via this skill are co-authored interactively, not solely AI-generated.
 
-```markdown
----
-provenance:
-  skill: plan
-  timestamp: '{ISO 8601 UTC timestamp}'
-  baseSha: '{short SHA of origin/main, omit if unresolvable}'
-  isInteractive: true
-ticket_id: '{ticket ID from session context, omit if null}'
-ticket_ref: '{ticket display ref, omit if null}'
-branch: '{branch name from session context}'
-commit: '{short hash of HEAD}'
-pr: '{full PR URL, omit if not resolved}'
-run_id: '{run id, omit when not in an orchestrated run}'
----
+The body following the frontmatter has this structure:
 
+```markdown
 # Plan: {Descriptive title}
 
 **Date**: {YYYY-MM-DD HH:MM UTC}
@@ -77,11 +65,9 @@ Resolve artifact directory based on context.
 
 The artifact frontmatter conforms to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema.
 
-<!-- include: ../../_partials/frontmatter-via-script.md -->
+Run `resolve-frontmatter.sh --skill plan --interactive true` via Bash. Prepend the output verbatim to the artifact body.
 
-- `provenance.skill`: always `plan`.
-- `provenance.isInteractive`: always `true`.
-<!-- /include -->
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ### Run context
 
diff --git a/packages/agents/content/skills/refine-plan/SKILL.md b/packages/agents/content/skills/refine-plan/SKILL.md
index c58c08f5..8e3505aa 100644
--- a/packages/agents/content/skills/refine-plan/SKILL.md
+++ b/packages/agents/content/skills/refine-plan/SKILL.md
@@ -151,11 +151,13 @@ Stamp the revised plan with frontmatter conforming to the [universal artifact fr
 
 The stamp writes the full canonical schema in one atomic write: the `provenance:` block plus the top-level canonical fields. The top-level fields come from the script; the `provenance:` block is computed from `{input-provenance}` plus the stamping logic below.
 
-<!-- include: ../../_partials/frontmatter-via-script.md -->
+This site uses `--format json` because the `provenance:` Block is case-branched on the input artifact's existing provenance — see [artifact-conventions.md](../_data/artifact-conventions.md#bespoke-frontmatter-composition).
 
-The `provenance:` block is **not** populated from the script. Construct it manually per the case branches below.
+Run `resolve-frontmatter.sh --format json` via Bash. It emits a JSON object with the universal artifact fields (`branch`, `commit`, `baseSha`, `pr`, `ticket_id`, `ticket_ref`, `platform`, `timestamp`, `run_id`). Use those values verbatim for the matching YAML keys. Optional fields the script omits from its output (`baseSha`, `pr`, `ticket_id`, `ticket_ref`, `run_id`) must be omitted from the frontmatter too — do not emit `null` or empty strings.
+
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
-<!-- /include -->
+The `provenance:` block is **not** populated from the script. Construct it manually per the case branches below.
 
 **Round-trip preservation:** the top-level canonical fields (`branch`, `commit`, `pr`, etc.) are always re-resolved from current session context via the script — they are not carried forward from `{input-provenance}`. This is correct: the output is a new artifact at a new point in time on a potentially different branch. The `provenance:` block's camelCase convention (`baseSha`, `isInteractive`, `refinedBy`) is preserved as-is on both read and write — there is no rename. Provenance fields carried forward from the input are `skill`, `baseSha`, `isInteractive`, and `iteration` (per the case branches).
 
diff --git a/packages/agents/content/skills/respond-to-review/SKILL.md b/packages/agents/content/skills/respond-to-review/SKILL.md
index 9e31e40d..2bd209a1 100644
--- a/packages/agents/content/skills/respond-to-review/SKILL.md
+++ b/packages/agents/content/skills/respond-to-review/SKILL.md
@@ -32,13 +32,21 @@ This skill bridges the gap between receiving a code review and implementing fixe
 
 The artifact's frontmatter conforms to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema.
 
-<!-- include: ../../_partials/frontmatter-via-script.md -->
+Source `$MODEL_ID` from your system-prompt environment block: the line `model named ... model ID is ...`. Set `$review_filename` to the bare filename of the review being responded to (e.g., `09_reviewer_review.md`).
 
-- `provenance.skill`: always `respond-to-review`.
-- `provenance.isInteractive`: always `true`.
-- `provenance.model`: the model identifier you are executing under. Read this from your system-prompt environment block — the line `model named ... model ID is ...`.
-- `responding_to` (response-artifact extension): the bare filename of the review being responded to (e.g., `09_reviewer_review.md`).
-<!-- /include -->
+Run via Bash:
+
+```bash
+resolve-frontmatter.sh \
+  --skill respond-to-review \
+  --interactive true \
+  --model "$MODEL_ID" \
+  --extra "responding_to=$review_filename"
+```
+
+Prepend the script's output verbatim to the artifact body.
+
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ## Locating the review
 
@@ -126,24 +134,11 @@ Per [artifact conventions](../_data/artifact-conventions.md#disposition-rules):
 
 When `ticket_ref` is null (no ticket on the branch), omit the `{ticket_ref}: ` portion so the heading reads `# Change summary: {description}`.
 
-The artifact begins with YAML frontmatter conforming to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema. The `responding_to` field is a response-artifact-specific extension that records the review being addressed.
+The artifact begins with YAML frontmatter conforming to the canonical schema; see the canonical example in [artifact-conventions.md](../_data/artifact-conventions.md#universal-artifact-frontmatter) and the field-resolution steps in the [Frontmatter resolution](#frontmatter-resolution) section above. The `responding_to` field is a response-artifact-specific extension that records the review being addressed.
 
-```markdown
----
-provenance:
-  skill: respond-to-review
-  timestamp: '{ISO 8601 UTC timestamp}'
-  baseSha: '{short SHA of origin/main, omit if unresolvable}'
-  isInteractive: true
-  model: '{model id}'
-ticket_id: '{ticket ID from session context, omit if null}'
-ticket_ref: '{ticket display ref, omit if null}'
-branch: '{branch name from session context}'
-commit: '{short hash of HEAD}'
-pr: '{full PR URL, omit if not resolved}'
-responding_to: '{filename of the review being responded to}'
----
+The body following the frontmatter has this structure:
 
+```markdown
 # Change summary: {ticket_ref}: {description}
 
 ## Changes made
diff --git a/packages/agents/content/skills/review-branch/SKILL.md b/packages/agents/content/skills/review-branch/SKILL.md
index aba7aff6..a9737c3e 100644
--- a/packages/agents/content/skills/review-branch/SKILL.md
+++ b/packages/agents/content/skills/review-branch/SKILL.md
@@ -42,12 +42,21 @@ This skill is the canonical home of the shared review process. `review-pr` invok
 
 The artifact's frontmatter conforms to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema.
 
-<!-- include: ../../_partials/frontmatter-via-script.md -->
+Source `$MODEL_ID` from your system-prompt environment block: the line `model named ... model ID is ...`. Resolve `$author` from `git log --format='%an' "$default_branch..HEAD" | sort -u | paste -sd, -` (unique authors of the commits under review).
 
-- `provenance.skill`: always `review-branch`.
-- `provenance.isInteractive`: always `true`.
-- `provenance.model`: the model identifier you are executing under. Read this from your system-prompt environment block — the line `model named ... model ID is ...`.
-<!-- /include -->
+Run via Bash:
+
+```bash
+resolve-frontmatter.sh \
+  --skill review-branch \
+  --interactive true \
+  --model "$MODEL_ID" \
+  --extra "author=$author"
+```
+
+Prepend the script's output verbatim to the artifact body.
+
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ## Review guidelines
 
@@ -78,24 +87,11 @@ Section-header icons (🚨, ⚠️, 📋, 🧠, ☝️, 🔍) come from the cano
 
 When `ticket_ref` is null (no ticket on the branch), omit the `{ticket_ref}: ` portion so the heading reads naturally without it — e.g., `# Code review: {description}`.
 
-The artifact begins with YAML frontmatter conforming to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema:
+The artifact begins with YAML frontmatter conforming to the canonical schema; see the canonical example in [artifact-conventions.md](../_data/artifact-conventions.md#universal-artifact-frontmatter) and the field-resolution steps in the [Frontmatter resolution](#frontmatter-resolution) section above. Pass `--extra "author=$author"` to the script to populate the review-artifact `author` field.
 
-```markdown
----
-provenance:
-  skill: review-branch
-  timestamp: '{ISO 8601 UTC timestamp}'
-  baseSha: '{short SHA of origin/main, omit if unresolvable}'
-  isInteractive: true
-  model: '{model id}'
-ticket_id: '{ticket ID from session context, omit if null}'
-ticket_ref: '{ticket display ref, omit if null}'
-branch: '{branch name from session context}'
-commit: '{short hash of HEAD}'
-pr: '{full PR URL, omit if not resolved}'
-author: '{commit author(s)}'
----
+The body following the frontmatter has this structure:
 
+```markdown
 # Code review: {ticket_ref}: {description in imperative mood}
 
 ## Summary of changes
diff --git a/packages/agents/content/skills/save-plan/SKILL.md b/packages/agents/content/skills/save-plan/SKILL.md
index d5e3aaf6..41b2d1ec 100644
--- a/packages/agents/content/skills/save-plan/SKILL.md
+++ b/packages/agents/content/skills/save-plan/SKILL.md
@@ -22,10 +22,9 @@ Save the plan from the current conversation as a ticket-scoped artifact. Useful
 
    The frontmatter conforms to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema plus the [plan provenance](../_data/artifact-conventions.md#plan-provenance) extensions.
 
-   <!-- include: ../../_partials/frontmatter-via-script.md -->
-   - `provenance.skill`: always `plan-mode`.
-   - `provenance.isInteractive`: always `true`.
-   <!-- /include -->
+   Run `resolve-frontmatter.sh --skill plan-mode --interactive true` via Bash. Prepend the output verbatim to the artifact body.
+
+   If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 5. **Save** as ticket-level artifact:
 
diff --git a/packages/agents/content/skills/summarize-change/SKILL.md b/packages/agents/content/skills/summarize-change/SKILL.md
index d022ce15..7bfcaced 100644
--- a/packages/agents/content/skills/summarize-change/SKILL.md
+++ b/packages/agents/content/skills/summarize-change/SKILL.md
@@ -32,27 +32,11 @@ If expected information is missing, stop and ask the developer.
 
 ## Output format
 
-The artifact begins with a single YAML frontmatter block that unifies canonical fields from the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) with change-summary-specific consumer fields documented at [Change-summary frontmatter](../_data/artifact-conventions.md#change-summary-frontmatter). Ordering: `provenance:` first, then top-level canonical fields, then consumer fields. `commit:` and `ticket_id:` appear exactly once each.
+The artifact begins with a single YAML frontmatter block that unifies canonical fields from the canonical schema with change-summary-specific consumer fields; see the canonical example in [artifact-conventions.md](../_data/artifact-conventions.md#universal-artifact-frontmatter) and the consumer-field extensions in [Change-summary frontmatter](../_data/artifact-conventions.md#change-summary-frontmatter). Ordering: `provenance:` first, then top-level canonical fields, then consumer fields. `commit:` and `ticket_id:` appear exactly once each. Field-resolution steps live in the [Canonical-field resolution](#canonical-field-resolution) section below.
 
-```markdown
----
-provenance:
-  skill: summarize-change
-  timestamp: '{ISO 8601 UTC timestamp}'
-  baseSha: '{short SHA of origin/main, omit if unresolvable}'
-  isInteractive: true
-  model: '{model id}'
-branch: '{branch name from session context}'
-commit: '{short hash of HEAD}'
-pr: '{full PR URL, omit if not resolved}'
-ticket_id: '{ticket ID from session context, omit if null}'
-ticket_ref: '{ticket display ref, omit if null}'
-run_id: '{run id, omit when not in an orchestrated run}'
-title: '{bare title without `ticket_ref` prefix}'
-scope: '{scope inferred from commit prefixes, or omitted if ambiguous}'
-type: '{work type inferred from commit prefixes, or omitted if ambiguous}'
----
+The body following the frontmatter has this structure:
 
+```markdown
 # {ticket_ref} {title}
 
 ## What
@@ -115,11 +99,25 @@ The block is structured as:
 
 ### Canonical-field resolution
 
-<!-- include: ../../_partials/frontmatter-via-script.md -->
+Source `$MODEL_ID` from your system-prompt environment block: the line `model named ... model ID is ...`. Resolve `$title`, `$scope`, and `$type` from the consumer-field inference below (omit a flag if the corresponding value cannot be inferred unambiguously).
+
+Run via Bash:
+
+```bash
+resolve-frontmatter.sh \
+  --skill summarize-change \
+  --interactive true \
+  --model "$MODEL_ID" \
+  --extra "title=$title" \
+  ${scope:+--extra "scope=$scope"} \
+  ${type:+--extra "type=$type"}
+```
+
+The `${var:+--extra "key=$var"}` form expands to the flag only when `$var` is non-empty, so an unresolved `scope` or `type` is naturally omitted from the emitted frontmatter.
+
+Prepend the script's output verbatim to the artifact body.
 
-- `provenance.skill`: always `summarize-change`.
-- `provenance.isInteractive`: always `true`.
-<!-- /include -->
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ### Consumer-field inference
 
diff --git a/packages/agents/content/skills/summarize-chat/SKILL.md b/packages/agents/content/skills/summarize-chat/SKILL.md
index 28f72e88..8bdff4f1 100644
--- a/packages/agents/content/skills/summarize-chat/SKILL.md
+++ b/packages/agents/content/skills/summarize-chat/SKILL.md
@@ -23,23 +23,11 @@ Replace `/Users/{username}/` with `~/` in file paths. Remove similar personal in
 
 ## Output format
 
-The artifact begins with YAML frontmatter conforming to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema. See [Frontmatter resolution](#frontmatter-resolution) below for field resolution.
+The artifact begins with YAML frontmatter conforming to the canonical schema; see the canonical example in [artifact-conventions.md](../_data/artifact-conventions.md#universal-artifact-frontmatter) and the field-resolution steps in the [Frontmatter resolution](#frontmatter-resolution) section below.
 
-```markdown
----
-provenance:
-  skill: summarize-chat
-  timestamp: '{ISO 8601 UTC timestamp}'
-  baseSha: '{short SHA of origin/main, omit if unresolvable}'
-  isInteractive: true
-  model: '{model id}'
-ticket_id: '{ticket id, omit if absent}'
-ticket_ref: '{ticket display ref, omit if absent}'
-branch: '{current branch name}'
-commit: '{short hash of HEAD}'
-pr: '{full PR URL, omit if not resolved}'
----
+The body following the frontmatter has this structure:
 
+```markdown
 # {Descriptive title}
 
 ## Problem
@@ -83,12 +71,11 @@ Use these to mark significant sections:
 
 The artifact's frontmatter conforms to the [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) schema.
 
-<!-- include: ../../_partials/frontmatter-via-script.md -->
+Source `$MODEL_ID` from your system-prompt environment block: the line `model named ... model ID is ...`.
+
+Run `resolve-frontmatter.sh --skill summarize-chat --interactive true --model "$MODEL_ID"` via Bash. Prepend the output verbatim to the artifact body.
 
-- `provenance.skill`: always `summarize-chat`.
-- `provenance.isInteractive`: always `true`.
-- `provenance.model`: the model identifier you are executing under. Read this from your system-prompt environment block — the line `model named ... model ID is ...`.
-<!-- /include -->
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ## Saving
 
diff --git a/packages/agents/content/skills/wrap-up/SKILL.md b/packages/agents/content/skills/wrap-up/SKILL.md
index 79a9d51b..be849b08 100644
--- a/packages/agents/content/skills/wrap-up/SKILL.md
+++ b/packages/agents/content/skills/wrap-up/SKILL.md
@@ -331,14 +331,19 @@ Prepend YAML frontmatter, then the markdown body.
 
 **Frontmatter** — see [universal artifact frontmatter](../_data/artifact-conventions.md#universal-artifact-frontmatter) for the canonical schema and [Deferred-findings frontmatter](../_data/artifact-conventions.md#deferred-findings-frontmatter) for the artifact-specific extensions.
 
-<!-- include: ../../_partials/frontmatter-via-script.md -->
+This site uses `--format json` because `tickets_created` is a list-of-objects extension that has no clean CLI expression; see [artifact-conventions.md](../_data/artifact-conventions.md#bespoke-frontmatter-composition).
+
+Run `resolve-frontmatter.sh --format json` via Bash. It emits a JSON object with the universal artifact fields (`branch`, `commit`, `baseSha`, `pr`, `ticket_id`, `ticket_ref`, `platform`, `timestamp`, `run_id`). Use those values verbatim for the matching YAML keys. Optional fields the script omits from its output (`baseSha`, `pr`, `ticket_id`, `ticket_ref`, `run_id`) must be omitted from the frontmatter too — do not emit `null` or empty strings.
+
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
+
+Set these skill-specific values inline (not in the script's output):
 
 - `provenance.skill`: always `wrap-up`.
 - `provenance.isInteractive`: always `true`.
 - `run_id`: **override** the script's value — reuse the run ID Phase 1a captured (also passed to `/create-devlog --run-id` in Phase 3). Emit only when wrap-up was invoked from an orchestrated session.
 - `session_type` (deferred-findings extension): the classification produced by Phase 1a's session-type detection (`orchestrated`, `interactive-dev`, `review`, or `research`).
 - `tickets_created` (deferred-findings extension): list of `{id, items}` entries cross-referencing each created ticket to the wrap-up item IDs it addresses. `items` is always a list. Omit when empty.
-<!-- /include -->
 
 **Body** — emit the tickets-created cross-reference and the dropped-findings record:
 
diff --git a/packages/agents/content/subagents/aspect-code-reviewer.md b/packages/agents/content/subagents/aspect-code-reviewer.md
index 64b8c359..f37dcce7 100644
--- a/packages/agents/content/subagents/aspect-code-reviewer.md
+++ b/packages/agents/content/subagents/aspect-code-reviewer.md
@@ -69,12 +69,11 @@ If the review concluded with no findings, the finalized form omits the `### Find
 
 The artifact's frontmatter conforms to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc).
 
-<!-- include: ../_partials/frontmatter-via-script.md -->
+Source `$MODEL_ID` from your system-prompt environment block: the line `model named ... model ID is ...`.
 
-- `provenance.skill`: always `aspect-code-reviewer`.
-- `provenance.isInteractive`: always `false`.
-- `provenance.model`: the model identifier you are executing under. Read this from your system-prompt environment block — the line `model named ... model ID is ...`.
-<!-- /include -->
+Run `resolve-frontmatter.sh --skill aspect-code-reviewer --interactive false --model "$MODEL_ID"` via Bash. Prepend the output verbatim to the artifact body.
+
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ## Scope
 
diff --git a/packages/agents/content/subagents/aspect-silent-failure-reviewer.md b/packages/agents/content/subagents/aspect-silent-failure-reviewer.md
index 586b5b30..f9512f28 100644
--- a/packages/agents/content/subagents/aspect-silent-failure-reviewer.md
+++ b/packages/agents/content/subagents/aspect-silent-failure-reviewer.md
@@ -63,12 +63,11 @@ If the review concluded with no findings (or no error-handling code was present)
 
 The artifact's frontmatter conforms to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc).
 
-<!-- include: ../_partials/frontmatter-via-script.md -->
+Source `$MODEL_ID` from your system-prompt environment block: the line `model named ... model ID is ...`.
 
-- `provenance.skill`: always `aspect-silent-failure-reviewer`.
-- `provenance.isInteractive`: always `false`.
-- `provenance.model`: the model identifier you are executing under. Read this from your system-prompt environment block — the line `model named ... model ID is ...`.
-<!-- /include -->
+Run `resolve-frontmatter.sh --skill aspect-silent-failure-reviewer --interactive false --model "$MODEL_ID"` via Bash. Prepend the output verbatim to the artifact body.
+
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ## Scope
 
diff --git a/packages/agents/content/subagents/aspect-test-reviewer.md b/packages/agents/content/subagents/aspect-test-reviewer.md
index 0597f810..bcb9751d 100644
--- a/packages/agents/content/subagents/aspect-test-reviewer.md
+++ b/packages/agents/content/subagents/aspect-test-reviewer.md
@@ -72,12 +72,11 @@ If the review concluded with no findings (or no source files required test cover
 
 The artifact's frontmatter conforms to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc).
 
-<!-- include: ../_partials/frontmatter-via-script.md -->
+Source `$MODEL_ID` from your system-prompt environment block: the line `model named ... model ID is ...`.
 
-- `provenance.skill`: always `aspect-test-reviewer`.
-- `provenance.isInteractive`: always `false`.
-- `provenance.model`: the model identifier you are executing under. Read this from your system-prompt environment block — the line `model named ... model ID is ...`.
-<!-- /include -->
+Run `resolve-frontmatter.sh --skill aspect-test-reviewer --interactive false --model "$MODEL_ID"` via Bash. Prepend the output verbatim to the artifact body.
+
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ## Scope
 
diff --git a/packages/agents/content/subagents/code-simplification-reviewer.md b/packages/agents/content/subagents/code-simplification-reviewer.md
index def4d65f..17478abb 100644
--- a/packages/agents/content/subagents/code-simplification-reviewer.md
+++ b/packages/agents/content/subagents/code-simplification-reviewer.md
@@ -64,12 +64,11 @@ If the review concluded with no findings, the finalized form omits the `### Find
 
 The artifact's frontmatter conforms to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc).
 
-<!-- include: ../_partials/frontmatter-via-script.md -->
+Source `$MODEL_ID` from your system-prompt environment block: the line `model named ... model ID is ...`.
 
-- `provenance.skill`: always `code-simplification-reviewer`.
-- `provenance.isInteractive`: always `false`.
-- `provenance.model`: the model identifier you are executing under. Read this from your system-prompt environment block — the line `model named ... model ID is ...`.
-<!-- /include -->
+Run `resolve-frontmatter.sh --skill code-simplification-reviewer --interactive false --model "$MODEL_ID"` via Bash. Prepend the output verbatim to the artifact body.
+
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ## Scope
 
diff --git a/packages/agents/content/subagents/orchestrated-architect.md b/packages/agents/content/subagents/orchestrated-architect.md
index 4c3fd48a..36988cf2 100644
--- a/packages/agents/content/subagents/orchestrated-architect.md
+++ b/packages/agents/content/subagents/orchestrated-architect.md
@@ -58,24 +58,9 @@ Classify the task into exactly one impact level:
 
 Write your analysis to the file path provided in your task prompt using the {tool:Write} tool. The artifact begins with YAML frontmatter conforming to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc) (see [Frontmatter](#frontmatter) below for field resolution).
 
-The document MUST include:
+The body following the frontmatter MUST include:
 
 ```markdown
----
-provenance:
-  skill: orchestrated-architect
-  timestamp: '{ISO 8601 UTC timestamp}'
-  baseSha: '{short SHA of origin/main, omit if unresolvable}'
-  isInteractive: false
-  model: '{model id}'
-ticket_id: '{ticket id, omit if absent}'
-ticket_ref: '{ticket display ref, omit if absent}'
-branch: '{current branch name}'
-commit: '{short hash of HEAD}'
-pr: '{full PR URL, omit if not resolved}'
-run_id: '{run id}'
----
-
 ### Impact level: {none|low|medium|high}
 
 ### Summary
@@ -128,12 +113,11 @@ If the plan's assumptions all check out, omit this section.
 
 The artifact's frontmatter conforms to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc).
 
-<!-- include: ../_partials/frontmatter-via-script.md -->
+Source `$MODEL_ID` from your system-prompt environment block: the line `model named ... model ID is ...`.
+
+Run `resolve-frontmatter.sh --skill orchestrated-architect --interactive false --model "$MODEL_ID"` via Bash. Prepend the output verbatim to the artifact body.
 
-- `provenance.skill`: Always `orchestrated-architect`.
-- `provenance.isInteractive`: Always `false`.
-- `provenance.model`: The model identifier you are executing under. Read this from your system-prompt environment block — the line `model named ... model ID is ...`.
-<!-- /include -->
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ## Principles
 
diff --git a/packages/agents/content/subagents/orchestrated-coder.md b/packages/agents/content/subagents/orchestrated-coder.md
index 3600951a..d760f8d2 100644
--- a/packages/agents/content/subagents/orchestrated-coder.md
+++ b/packages/agents/content/subagents/orchestrated-coder.md
@@ -137,12 +137,11 @@ completed
 
 The artifact's frontmatter conforms to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc).
 
-<!-- include: ../_partials/frontmatter-via-script.md -->
+Source `$MODEL_ID` from your system-prompt environment block: the line `model named ... model ID is ...`.
 
-- `provenance.skill`: always `orchestrated-coder`.
-- `provenance.isInteractive`: always `false`.
-- `provenance.model`: the model identifier you are executing under. Read this from your system-prompt environment block — the line `model named ... model ID is ...`.
-<!-- /include -->
+Run `resolve-frontmatter.sh --skill orchestrated-coder --interactive false --model "$MODEL_ID"` via Bash. Prepend the output verbatim to the artifact body.
+
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ## Reviewer-context sidecar
 
diff --git a/packages/agents/content/subagents/orchestrated-planner.md b/packages/agents/content/subagents/orchestrated-planner.md
index e0c7fd2e..e17e8184 100644
--- a/packages/agents/content/subagents/orchestrated-planner.md
+++ b/packages/agents/content/subagents/orchestrated-planner.md
@@ -44,24 +44,9 @@ You will receive:
 
 ## Output: Plan (Markdown)
 
-Write the plan Markdown file to the path provided in the task prompt. The artifact begins with YAML frontmatter conforming to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc) (see [Frontmatter](#frontmatter) below for field resolution). Format:
+Write the plan Markdown file to the path provided in the task prompt. The artifact begins with YAML frontmatter conforming to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc) (see [Frontmatter](#frontmatter) below for field resolution). The frontmatter conforms to the canonical schema; see the canonical example in the `artifact-conventions` data doc. Format:
 
 ```markdown
----
-provenance:
-  skill: orchestrated-planner
-  timestamp: '{ISO 8601 UTC timestamp}'
-  baseSha: '{short SHA of origin/main, omit if unresolvable}'
-  isInteractive: false
-  model: '{model id}'
-ticket_id: '{ticket id, omit if absent}'
-ticket_ref: '{ticket display ref, omit if absent}'
-branch: '{current branch name}'
-commit: '{short hash of HEAD}'
-pr: '{full PR URL, omit if not resolved}'
-run_id: '{run id}'
----
-
 # Implementation Plan
 
 ## Overview
@@ -148,12 +133,11 @@ Write the plan JSON file to the path provided in the task prompt. Format:
 
 The artifact's frontmatter conforms to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc).
 
-<!-- include: ../_partials/frontmatter-via-script.md -->
+Source `$MODEL_ID` from your system-prompt environment block: the line `model named ... model ID is ...`.
+
+Run `resolve-frontmatter.sh --skill orchestrated-planner --interactive false --model "$MODEL_ID"` via Bash. Prepend the output verbatim to the artifact body.
 
-- `provenance.skill`: Always `orchestrated-planner`.
-- `provenance.isInteractive`: Always `false`.
-- `provenance.model`: The model identifier you are executing under. Read this from your system-prompt environment block — the line `model named ... model ID is ...`.
-<!-- /include -->
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ## Constraints
 
diff --git a/packages/agents/content/subagents/orchestrated-reviewer.md b/packages/agents/content/subagents/orchestrated-reviewer.md
index fd93a5c3..e95f3ecd 100644
--- a/packages/agents/content/subagents/orchestrated-reviewer.md
+++ b/packages/agents/content/subagents/orchestrated-reviewer.md
@@ -73,12 +73,11 @@ If the review concluded with no findings, the finalized form omits the `### Find
 
 The artifact's frontmatter conforms to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc).
 
-<!-- include: ../_partials/frontmatter-via-script.md -->
+Source `$MODEL_ID` from your system-prompt environment block: the line `model named ... model ID is ...`.
 
-- `provenance.skill`: always `orchestrated-reviewer`.
-- `provenance.isInteractive`: always `false`.
-- `provenance.model`: the model identifier you are executing under. Read this from your system-prompt environment block — the line `model named ... model ID is ...`.
-<!-- /include -->
+Run `resolve-frontmatter.sh --skill orchestrated-reviewer --interactive false --model "$MODEL_ID"` via Bash. Prepend the output verbatim to the artifact body.
+
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ## Finding format
 
diff --git a/packages/agents/content/subagents/plan-reviewer.md b/packages/agents/content/subagents/plan-reviewer.md
index c9af000d..1895395b 100644
--- a/packages/agents/content/subagents/plan-reviewer.md
+++ b/packages/agents/content/subagents/plan-reviewer.md
@@ -65,26 +65,11 @@ Each finding is tagged with a resolution type:
 
 ## Output format
 
-Write the review to the output path provided in your task prompt. The artifact begins with YAML frontmatter conforming to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc) (see [Frontmatter](#frontmatter) below for field resolution).
+Write the review to the output path provided in your task prompt. The artifact begins with YAML frontmatter conforming to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc); see the [Frontmatter](#frontmatter) section below for field resolution.
 
 **Section organization:** Sections are grouped by **resolution type** (auto vs user), not by finding category (C vs X). Place every `auto`-tagged finding -- whether C or X -- in "Auto-resolvable findings". Place every `user`-tagged finding -- whether C or X -- in "Decision gaps". Every finding in "Decision gaps" must include a **Question** field.
 
 ```markdown
----
-provenance:
-  skill: plan-reviewer
-  timestamp: '{ISO 8601 UTC timestamp}'
-  baseSha: '{short SHA of origin/main, omit if unresolvable}'
-  isInteractive: false
-  model: '{model id}'
-ticket_id: '{ticket id, omit if absent}'
-ticket_ref: '{ticket display ref, omit if absent}'
-branch: '{current branch name}'
-commit: '{short hash of HEAD}'
-pr: '{full PR URL, omit if not resolved}'
-run_id: '{run id, omit if not in a run}'
----
-
 # Plan review
 
 ## Summary
@@ -152,12 +137,11 @@ If the plan has no findings at all, write:
 
 The artifact's frontmatter conforms to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc).
 
-<!-- include: ../_partials/frontmatter-via-script.md -->
+Source `$MODEL_ID` from your system-prompt environment block: the line `model named ... model ID is ...`.
+
+Run `resolve-frontmatter.sh --skill plan-reviewer --interactive false --model "$MODEL_ID"` via Bash. Prepend the output verbatim to the artifact body.
 
-- `provenance.skill`: Always `plan-reviewer`.
-- `provenance.isInteractive`: Always `false`.
-- `provenance.model`: The model identifier you are executing under. Read this from your system-prompt environment block — the line `model named ... model ID is ...`.
-<!-- /include -->
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ## Principles
 
diff --git a/packages/agents/content/subagents/planner.md b/packages/agents/content/subagents/planner.md
index f12ce01a..333b7262 100644
--- a/packages/agents/content/subagents/planner.md
+++ b/packages/agents/content/subagents/planner.md
@@ -74,24 +74,9 @@ Write the machine-readable plan to `{plan-json-path}`:
 
 ## Output: orchestration-plan.md
 
-Write the human-readable plan to `{plan-md-path}`. The artifact begins with YAML frontmatter conforming to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc) (see [Frontmatter](#frontmatter) below for field resolution). The JSON sidecar does not carry frontmatter.
+Write the human-readable plan to `{plan-md-path}`. The artifact begins with YAML frontmatter conforming to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc) (see [Frontmatter](#frontmatter) below for field resolution). The JSON sidecar does not carry frontmatter. The frontmatter conforms to the canonical schema; see the canonical example in the `artifact-conventions` data doc.
 
 ```markdown
----
-provenance:
-  skill: planner
-  timestamp: '{ISO 8601 UTC timestamp}'
-  baseSha: '{short SHA of origin/main, omit if unresolvable}'
-  isInteractive: false
-  model: '{model id}'
-ticket_id: '{ticket id, omit if absent}'
-ticket_ref: '{ticket display ref, omit if absent}'
-branch: '{current branch name}'
-commit: '{short hash of HEAD}'
-pr: '{full PR URL, omit if not resolved}'
-run_id: '{run id, omit if not in a run}'
----
-
 # Implementation Plan
 
 ## Overview
@@ -138,12 +123,11 @@ run_id: '{run id, omit if not in a run}'
 
 The artifact's frontmatter conforms to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc).
 
-<!-- include: ../_partials/frontmatter-via-script.md -->
+Source `$MODEL_ID` from your system-prompt environment block: the line `model named ... model ID is ...`.
+
+Run `resolve-frontmatter.sh --skill planner --interactive false --model "$MODEL_ID"` via Bash. Prepend the output verbatim to the artifact body.
 
-- `provenance.skill`: Always `planner`.
-- `provenance.isInteractive`: Always `false`.
-- `provenance.model`: The model identifier you are executing under. Read this from your system-prompt environment block — the line `model named ... model ID is ...`.
-<!-- /include -->
+If the script's stderr contains `Note: PR lookup failed; proceeding without pr field.`, surface that line in your text output once.
 
 ## Resumption
 
diff --git a/packages/agents/content/subagents/savings-analyzer.md b/packages/agents/content/subagents/savings-analyzer.md
index 6ce0b8cb..27939ad2 100644
--- a/packages/agents/content/subagents/savings-analyzer.md
+++ b/packages/agents/content/subagents/savings-analyzer.md
@@ -64,24 +64,11 @@ Tag every suggestion:
 
 ## Output format
 
-Write a markdown artifact with this structure. The artifact begins with YAML frontmatter conforming to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc) (see [Frontmatter](#frontmatter) below for field resolution).
+Write a markdown artifact with this structure. The artifact begins with YAML frontmatter conforming to the universal artifact frontmatter schema (defined in the `artifact-conventions` shared data doc); see the [Frontmatter](#frontmatter) section below for field resolution.
 
-```
----
-provenance:
-  skill: savings-analyzer
-  timestamp: '{ISO 8601 UTC timestamp}'
-  baseSha: '{short SHA of origin/main, omit if unresolvable}'
-  isInteractive: false
-  model: '{model id}'
-ticket_id: '{ticket id, omit if absent}'
-ticket_ref: '{ticket display ref, omit if absent}'
-branch: '{current branch name}'
-commit: '{short hash of HEAD}'
-pr: '{full PR URL, omit if not resolved}'
-run_id: '{run id}'
----
+The body following the frontmatter MUST include:
 
+```
 # Savings analysis
 
 ## Summary
@@ -114,7 +101,7 @@ Resolve fields before writing the artifact:
 - `provenance.timestamp`: Current UTC time in ISO 8601 format.
 - `provenance.baseSha`: Passed in via your dispatch prompt — the orchestrator resolves `git rev-parse --short origin/main` for the run-summary and forwards it. Omit if not provided.
 - `provenance.isInteractive`: Always `false`.
-- `provenance.model`: The model identifier you are executing under. Read this from your system-prompt environment block — look for the line `model named ... model ID is ...` and use the model ID value.
+- `provenance.model`: The model identifier you are executing under. Read this from your system-prompt environment block: Look for the line `model named ... model ID is ...` and use the model ID value.
 - `ticket_id`, `ticket_ref`: Passed in via your dispatch prompt. Omit when absent.
 - `branch`: Passed in via your dispatch prompt.
 - `commit`: Passed in via your dispatch prompt — the short HEAD SHA at run time.