Add the persona @-mention router#754
Conversation
Serves the addressing contract that .github#752 defined. ONE router for every persona — personas do NOT each ship a mention workflow (§4.1); a stub per agent is exactly the drift the manifest exists to prevent, and it is how the fleet got two divergent registries the last time. There is no persona index, by design Because address.handle's slug MUST equal `id`, and `id` MUST equal the persona's directory name (both enforced by validate-personas.py), '@petry-projects/qa-lead' resolves to personas/qa-lead/persona.yml by convention. The manifest is the index-of-record (§1.1) — nothing derives, caches or duplicates it. This also avoids inverting the established derivation direction: canary-rings.json (.github) -> release/registry.yml (.github-private), whereas a persona index would have had to flow .github-private -> .github. A 404 is a meaningful answer ("not a persona"), which is also how a real, non-persona team like @petry-projects/org-leads falls through harmlessly. What decides what The manifest, not this router: whether the mention surface is enabled, in what mode, behind which trust floor, under which opt_out_label. The router restates none of it. A persona may TIGHTEN the job-level trust default but never loosen it — an undeclared floor denies, so a persona that forgot to declare trust is not more permissive than one that did. Recursion (the reason the core is a tested library, not inline shell) Comments posted via a PAT re-trigger workflows, unlike GITHUB_TOKEN. .github-private#860 burned 1,481 identical acks in 4.5h from a SINGLE self-loop. With N mutually addressable personas the cycles are combinatorial, not self-loops: qa-lead answering a thread that mentions dev-lead is enough. So the guard is two independent axes — bot actor AND agent marker — enforced both in the job-level `if` (so secrets never reach a bad run) and in pm_should_route (so it is testable). The marker match is a PREFIX: #860's first fix matched one exact ack string and still self-looped through a different agent comment. Contents - scripts/lib/persona-mention.sh — the pure, network-free decision core. Fetching is the workflow's job, so the suite pins routing behaviour with no API access at all. - tests/persona_mention.bats — 31 tests, recursion guards most heavily. - .github/workflows/persona-mention-tests.yml — the runner. A bats suite with no CI runner is unenforced (the #613/#624 gap canary-rollout-tests.yml exists to close); added WITH the suite rather than after it. - .github/workflows/persona-mention-reusable.yml — the router. Sources the lib at `github.job_workflow_sha` so the logic always matches the pinned workflow version (the #465/#528 tooling_ref lesson); every event field arrives via env, never inline `${{ }}` in a run: body. - standards/workflows/persona-mention.yml — the single caller stub, pinned to persona-mention/v1-stable (major-scoped per #657, matching live stubs). - standards/canary-rings.json — registers the reusable so a release can be cut. Deliberately NOT here - ring-pins.sh RING_REUSABLES is untouched: adding persona-mention would make compliance-audit demand the stub in repos that do not have it and make the deploy sweep report drift — manufacturing failures for a reusable with no cut release. That belongs to rollout. - No release cut, no repo enrolled, and no persona runtime exists yet, so NOTHING routes on merge. The router dispatches `persona-mention` to .github-private, where nothing listens yet — a safe no-op. Verification (CI's exact invocations, run locally) - bats tests/persona_mention.bats: 31/31. - shellcheck --severity=warning -x scripts/**/*.sh: clean. - actionlint -ignore SC2129 -ignore '"job_workflow_sha" is not defined' .github/workflows/*.yml: clean. (That ignore is pre-existing and correct — job_workflow_sha is a real context property actionlint 1.7.7's schema omits; the merged auto-rebase-reusable.yml trips the same false positive.) - markdownlint-cli2: clean. - canary-rings.json: spliced as TEXT, not re-serialized — the file's encoding is genuinely mixed (em-dashes escaped, arrows half-literal) and its agent order is insertion order, not alphabetical, so a json round-trip rewrote ~140 untouched lines. Result is 63 insertions / 0 deletions, and the entry is structurally identical to pr-review-mention's but for host/reusable/run_workflow. One bug the tests caught: pm_manifest_query piped python into jq, so the pipeline reported JQ's status and a parse failure returned empty output with exit 0 — indistinguishable from "the manifest says no". Now captured and checked. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
|
Warning Review limit reached
Next review available in: 59 minutes Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available. How can I continue?After more reviews become available, a review can be triggered using the To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews. How do review limits work?CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability. For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window. Please refer docs for additional details. Review details⚙️ Run configurationConfiguration used: Organization UI Review profile: ASSERTIVE Plan: Pro Run ID: 📒 Files selected for processing (2)
📝 WalkthroughWalkthroughAdds guarded persona-mention routing, reusable GitHub Actions execution, manifest-based trust and label handling, private repository dispatches, caller wiring, rollout registration, documentation, and automated Bats/ShellCheck validation. ChangesPersona mention routing
Estimated code review effort: 4 (Complex) | ~45 minutes Sequence Diagram(s)sequenceDiagram
participant Comment
participant CallerWorkflow
participant ReusableRouter
participant PersonaManifest
participant PrivateRepository
Comment->>CallerWorkflow: created issue or review comment
CallerWorkflow->>ReusableRouter: invoke reusable workflow
ReusableRouter->>PersonaManifest: fetch persona manifest
PersonaManifest-->>ReusableRouter: routing, labels, and trust settings
ReusableRouter->>PrivateRepository: dispatch persona-mention event
Possibly related issues
Possibly related PRs
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces a reusable Bash library and workflow for routing persona @-mention decisions, complete with recursion guards, manifest parsing, and BATS unit tests. The feedback recommends enhancing the robustness of the jq queries within the Bash library by incorporating the optional/try operator ? to prevent null-indexing errors, guarding against empty inputs, and expanding the test suite to cover these null-safety paths.
Dev-Lead — waiting on PR blockers (intent: review-changes)PR: #754 |
|
Note @don-petry I reviewed this PR and no code changes were needed, but it still has blocking checks or reviews (failing or cancelled checks, or changes-requested reviews), so I cannot mark it done yet. I'll re-check automatically. |
There was a problem hiding this comment.
Pull request overview
Adds the org-standard “persona mention router” that parses @petry-projects/<role> mentions, consults each persona’s manifest in .github-private, and dispatches to the persona runtime without per-persona workflow stubs (centralized routing per the persona addressing contract).
Changes:
- Introduces a pure, network-free routing decision library (
scripts/lib/persona-mention.sh) plus a bats test suite and CI runner to quality-gate it. - Adds a reusable workflow router (
persona-mention-reusable.yml) and the single caller stub template (standards/workflows/persona-mention.yml). - Registers the new reusable in
standards/canary-rings.jsonand documents/adverts the stub in the standards docs.
Reviewed changes
Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.
Show a summary per file
| File | Description |
|---|---|
scripts/lib/persona-mention.sh |
Adds the routing decision core (slug extraction, recursion guards, trust/manifest parsing). |
tests/persona_mention.bats |
Adds unit coverage (31 cases) emphasizing recursion and manifest/trust behaviors. |
.github/workflows/persona-mention-tests.yml |
Adds CI quality gate to run shellcheck + bats for the routing core. |
.github/workflows/persona-mention-reusable.yml |
Adds the reusable router that fetches manifests and dispatches persona runs. |
standards/workflows/persona-mention.yml |
Adds the thin caller stub template for consumer repos. |
standards/persona-standards.md |
Documents the router, its decision core, and the “no persona index” convention. |
standards/ci-standards.md |
Registers the new caller stub in the standards table of workflow templates. |
standards/canary-rings.json |
Registers the reusable in the canary channel/ring registry for future rollout. |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 5
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In @.github/workflows/persona-mention-reusable.yml:
- Around line 126-131: Update the label lookup block around LABELS so discussion
comments retain an empty label set, while issue and pull request lookups
propagate gh api failures instead of converting them to empty output. Remove the
fallback that masks errors and preserve the existing ITEM_NUMBER guard.
- Around line 138-143: Update the manifest fetch in the persona-processing
workflow to authenticate private `.github-private` requests using the existing
GH_TOKEN/PAT, either through the GitHub Contents API or an Authorization header
on the curl request. Preserve the current failure handling and manifest parsing
flow while ensuring valid private manifests are retrieved instead of treated as
missing.
In `@scripts/lib/persona-mention.sh`:
- Around line 169-185: Write-mode mention decisions currently omit the required
gate label, allowing unarmed items to dispatch. Update pm_mention_decision in
scripts/lib/persona-mention.sh to emit the mention surface’s gate_label
alongside enabled, mode, and opt_out_label; update tests/persona_mention.bats at
lines 188-194 to verify gate preservation and allowed/denied label states;
update .github/workflows/persona-mention-reusable.yml at lines 155-192 to block
write-mode dispatch unless the required gate label is present.
- Around line 81-105: The direct slug lookup in pm_manifest_url breaks the
existing persona alias and rename contract. Update the persona mention
resolution flow, including pm_extract_slugs and pm_manifest_url, to resolve each
addressed slug through the established alias mapping and construct the manifest
URL from the canonical persona ID; update standards/persona-standards.md lines
253-258 to document this alias-resolution mechanism, or explicitly remove
aliases from the addressing contract.
In `@standards/workflows/persona-mention.yml`:
- Around line 38-48: Subscribe to pull_request review_requested events in the
workflow trigger configuration in standards/workflows/persona-mention.yml. In
.github/workflows/persona-mention-reusable.yml, update the event normalization
logic to handle reviewer-assignment payloads by normalizing requested_team and
pull-request fields without assuming comment is present; preserve existing
comment-event handling.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro
Run ID: 5d4c8e2a-377a-4330-9d97-be017de4be50
📒 Files selected for processing (8)
.github/workflows/persona-mention-reusable.yml.github/workflows/persona-mention-tests.ymlscripts/lib/persona-mention.shstandards/canary-rings.jsonstandards/ci-standards.mdstandards/persona-standards.mdstandards/workflows/persona-mention.ymltests/persona_mention.bats
Dev-Lead — fix-reviews (applied)Changes committed and pushed. |
Dev-Lead — review-changes (no-changes)No changes were needed for this PR. |
Addresses the review findings on this PR (#755) and the bot review. 1. Fail-open trio — errors were read as negative answers The systemic flaw #755 names as its headline finding. Three instances here: - Manifest fetch: `curl -f` collapses 404 and 5xx into one exit code, so a raw.githubusercontent hiccup silently disabled EVERY persona fleet-wide, with no error anywhere and indistinguishable from nobody being addressed. Now switched on the status code: 200 routes, 404 is a real answer ("not a persona" — a plain team like @petry-projects/org-leads, or a typo), anything else is a hard failure. - Labels: `|| echo ""` turned an API error into "no labels", which would bypass an opt-out AND arm an ungated write. Now fails closed. Discussions genuinely have no labels surface, so an empty set is the TRUE answer there — that difference is load-bearing (see 2). - pm_manifest_query already fixed in 518ff2b (pipeline reported jq's status). 2. gate_label is now ENFORCED (security) The schema makes gate_label required for write surfaces, but only enforces that it is DECLARED — declaring a lock is not locking the door. The router read `mode`, passed it in the payload, and dispatched. '@petry-projects/dev-lead do X' would have been an ungated write the moment the first write persona onboarded. Now: write-mode requires the gate label to be APPLIED; a write persona with no gate_label is a hard error; and write-mode on a surface where labels are unavailable (discussions) is refused, because the gate cannot be verified. gate_label is read by its own function, not a 4th space-separated field on pm_mention_decision: an empty optional field silently shifts the next one into its place ("true write qa-lead" parses the gate AS the opt_out_label). A security gate is the last place to accept that. Test 42 pins it. 3. contents: read + PyYAML (Copilot) The job checks out the tooling repo but never granted contents: read. PyYAML is NOT documented as preinstalled on the runner images, and this reusable runs on the CALLER's runner, so it cannot be assumed — our own tests install it explicitly, which proved the gap. Added a cheap no-op-when-present step. Also persist-credentials: false on the tooling checkout. 4. Mention precision (#755 finding 9) The router fired where GitHub itself renders NO mention: fenced code, inline code. Pasting a usage example or documenting the handle summoned the persona. Blockquotes are a DELIBERATE divergence — GitHub does notify on a quoted mention, but its one-click "Quote reply" copies a whole prior comment prefixed '>', and re-running an agent over quoted text is not what the quoter meant. Neither recursion axis catches it (a human quoting a human is not a bot and carries no marker). To address a persona, mention it outside the quote. Verification (CI's exact invocations) - bats tests/persona_mention.bats: 42/42 (31 + 11 new covering fences, tilde fences, inline code, blockquotes, the two must-still-fire cases, and the gate_label read-shift trap). - shellcheck --severity=warning -x scripts/**/*.sh: clean. - actionlint (CI's ignores): clean. NOT fixed here, deliberately - aliases (#755 finding 1): the merged schema promises address.aliases[] but index-free routing cannot honour it. Dropping it needs an ORDERED cross-repo pair — .github-private must remove `aliases: []` from qa-lead's manifest first, or dropping it from the schema (additionalProperties: false) breaks validate-personas on main fleet-wide. Tracked in #755; it is a contract change and deserves its own PR. - review_requested routing (#755 finding 10) and the 👀 ack (finding 11) are scoped out to keep this PR about correctness and security. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The stub subscribed to pull_request/review_requested, but the router has no path for it — the job's `if` never admits the event, so every reviewer assignment fleet-wide would start a run that immediately skips. §4 does say "Reviewer-assignment counts here too", and it is a real gap (#755 finding 10). But the trigger lands WITH the routing, not before it: a trigger firing into a router that ignores it is dead weight that only burns runs and makes the stub lie about what it supports. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
Pushed the review fixes in FixedThe fail-open trio (#755's headline finding — errors read as negative answers):
Mention precision: no longer fires inside code fences or inline code, where GitHub itself renders no mention. Blockquotes are a deliberate divergence — GitHub does notify on quoted mentions, but one-click Quote reply copies a whole prior comment prefixed I dropped two things from
|
|
CI checks on this PR are still running. Once they complete, re-mention Posted by the donpetry-bot PR-review cascade. |
There was a problem hiding this comment.
Actionable comments posted: 2
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In @.github/workflows/persona-mention-reusable.yml:
- Around line 179-196: Update the manifest fetch in the curl invocation to send
the available GH_TOKEN as an Authorization header when requesting
.github-private URLs. Preserve the existing HTTP status handling, including
treating genuine 404 responses as non-personas and failing other fetch errors.
In `@standards/workflows/persona-mention.yml`:
- Around line 49-53: Fix the indentation of the explanatory comment beginning
“NOTE: reviewer assignment” in the workflow YAML so it matches the surrounding
block structure and satisfies yamllint’s comments-indentation rule. Preserve the
comment text and its placement.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro
Run ID: d8d07379-0dcf-407a-a62b-affec23c96b0
📒 Files selected for processing (5)
.github/workflows/persona-mention-reusable.ymlscripts/lib/persona-mention.shscripts/persona-mention-requirements.txtstandards/workflows/persona-mention.ymltests/persona_mention.bats
Dev-Lead — fix-reviews (applied)Changes committed and pushed. |
Dev-Lead — review-changes (no-changes)No changes were needed for this PR. |
…ent-death path Reverts the Authorization header added in 9333997, which was made in response to CodeRabbit's 🔴 Critical "the repo is private, so raw.githubusercontent will 404 and every persona gets skipped". That premise is false, and acting on it introduces the exact bug this PR exists to fix. petry-projects/.github-private is PUBLIC despite its name (private: false — every one of the org's 11 repos is public). Measured against the real manifest URL: unauthenticated -> 200 valid token -> 200 expired token -> 404 empty token -> 404 raw.githubusercontent answers a bad or MISSING token with 404, not 401. Our router — correctly — treats 404 as a real answer: "not a persona". So with an Authorization header, the moment GH_PAT_WORKFLOWS expires, or a repo adopts the stub before the secret exists, or a fork PR runs without secrets, EVERY persona 404s and is silently skipped as "not a persona". Fleet-wide, silent, and indistinguishable from nobody being addressed. Unauthenticated cannot fail that way for a public repo. The header buys nothing (200 either way) and adds a silent-death path — the precise fail-open class this PR closes. Reverted, with the measurements recorded inline so the next reader (human or agent) does not re-add it on the same false premise. The stub's comment re-indent from 9333997 is kept; it is cosmetic and fine. Verification - bats tests/persona_mention.bats: 42/42. - actionlint (CI's ignores): clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
don-petry
left a comment
don-petry
left a comment
There was a problem hiding this comment.
@coderabbitai your 🔴 Critical (".github-private is private, so raw.githubusercontent returns 404 and every persona gets skipped") is a false positive driven by the repo's name, and acting on it introduced the exact bug this PR exists to fix. Reverted in the latest commit, with the evidence recorded inline.
petry-projects/.github-private is PUBLIC — private: false. Every one of the org's 11 repos is public; the name is a leftover and lies.
Measured against the real manifest URL:
| Auth | Result |
|---|---|
| unauthenticated | 200 |
| valid token | 200 |
| expired token | 404 |
| empty token | 404 |
raw.githubusercontent answers a bad or missing token with 404, not 401. This router — correctly — treats 404 as a real answer: "not a persona".
So adding Authorization: Bearer $GH_TOKEN means that the moment GH_PAT_WORKFLOWS expires, or a repo adopts the stub before the secret exists, or a fork PR runs without secrets, every persona 404s and is silently skipped — fleet-wide, with no error, indistinguishable from nobody being addressed. That is precisely the fail-open class this PR closes (#755's headline finding).
Unauthenticated cannot fail that way for a public repo. The header buys nothing — 200 either way — and adds a silent-death path. The measurements are now inline above the case so the next reader doesn't re-add it on the same premise.
Dev-Lead — review-changes (no-changes)No changes were needed for this PR. |
|
Dev-Lead — review-changes (no-changes)No changes were needed for this PR. |
donpetry-bot
left a comment
donpetry-bot
left a comment
There was a problem hiding this comment.
Automated review — APPROVED ✓
Risk: MEDIUM
Reviewed commit: b022ed0fabb5eec1489f7620ba0e14b25554c977
Review mode: triage-approved (single reviewer)
Summary
Adds the org-wide persona @-mention router: a reusable workflow, a pure fail-closed routing library (scripts/lib/persona-mention.sh), a thin caller stub, 42 bats tests, canary-ring registration, and standards updates. Security posture is strong: all event fields reach shell via env (never inline ${{ }}), the trust floor is enforced in the job-level if: before secrets are exposed, actions are SHA-pinned, PyYAML is hash-locked, permissions default to {}, and recursion is guarded on two independent axes (bot actor + persona marker) per the #860 postmortem. Triage's low-risk call is broadly confirmed; I rate it MEDIUM (new workflow trust surface with an org PAT) — still within auto-approve range.
Linked issue analysis
No closing-issue references. The PR implements the addressing contract defined in #752 (persona-standards.md §4.1) and closes out findings carried from #755 (fail-open trio, write-mode gate arming, mention precision) — each is substantively addressed in the diff with matching regression tests.
Findings
Verified disputed finding — resolved in the author's favor. CodeRabbit's critical claim that petry-projects/.github-private is private (so unauthenticated raw fetches would 404 fleet-wide) is a false positive: the repo is public (verified via API: private=false, visibility=public). The unauthenticated fetch is the correct design — an Authorization header would convert an expired/missing PAT into silent 404s for every persona.
Review threads: all 15 resolved; CodeRabbit's two CHANGES_REQUESTED reviews are superseded (latest states all COMMENTED).
Non-blocking nits:
- scripts/lib/persona-mention.sh has a duplicated doc-comment block for pm_extract_slugs (the header appears above pm_strip_unaddressable and again above the function).
- In the reusable workflow, an unparseable manifest yields an empty pm_mention_decision result and is skipped as "not enabled" rather than erroring loudly (the here-string swallows the query's exit status). Fail-closed, so safe — but slightly at odds with the fail-loud philosophy applied elsewhere.
Secret scan: run_secret_scanning MCP tool unavailable in this run; gitleaks CI check passed, and the only hashes in the diff are pip package integrity hashes.
CI status
All checks green: CodeQL, SonarCloud (quality gate passed, 0 new issues), ShellCheck, bats, gitleaks, npm audit, AgentShield, Agent Security Scan, Lint. Ecosystem audits not applicable were skipped.
Reviewed automatically by the PR-review agent (single-reviewer mode: fable 5). Reply if you need a human review.
…thub-private#1289 merges (#769) * refactor(personas): drop address.aliases from the contract (half 2 of 2) Second half of the ordered cross-repo pair. petry-projects/.github-private#1289 removed every USE of the field first; this removes the field itself. MERGE ORDER: #1289 MUST land before this. The schema is additionalProperties:false, so dropping `aliases` while qa-lead's manifest still carries `aliases: []` fails validate-personas on main FLEET-WIDE. Verified the rejection is real: a manifest with `aliases` now yields "Additional properties are not allowed ('aliases' was unexpected)". Why aliases are gone (see .github#755 finding 1) They existed so a RENAMED role kept routing — qa-lead listing `petry-projects/murat` so old comments still reached it. Two reasons it goes: 1. Structural: index-free routing could never honour it. The router resolves '@petry-projects/murat' to personas/murat/persona.yml — a 404 after the rename. The alias is declared INSIDE the renamed persona's own manifest, which the router never opens, because it does not know to look there. Chicken-and-egg; the pm_persona_aliases attempt in #754 was unreachable for exactly this reason. 2. Measured: there is nothing live left to route. GitHub stops rendering a renamed team's old handle as a mention at all. Probed it directly — created a team, rendered its mention, renamed it, re-rendered the OLD handle: before rename: <a class="team-mention" …>@petry-projects/probe</a> after rename: <p>@petry-projects/probe please review</p> The old handle becomes PLAIN TEXT. So it is not a silently-broken mention that looks live — the reader sees grey text and knows instantly it did not resolve, and the fix is local: mention the new handle. Aliases would only route text GitHub itself no longer treats as a mention. Not worth a derived index, an extra fetch on every 404, and a contract the router cannot honour. Rule 4 becomes a theorem, not a check "Handles and aliases MUST be unique across all personas" is replaced by "Handles are unique BY CONSTRUCTION": the pattern pins the org, slug == id, id == the persona directory name, and directory names are unique — so two personas cannot claim the same handle. #1289 removed the corresponding check_handles_unique for the same reason: it could never fire. Rule 5 ("a rename keeps the old handle as an alias") is replaced by a §4.1 "Renames" section documenting what to change on a rename and stating plainly that old mentions stop resolving VISIBLY — which is the correct signal, not a regression. Changes - standards/personas/persona.schema.json — drop address.aliases (edited as TEXT, not re-serialized: a json round-trip reformats every inline array and rewrote 264 untouched lines. Diff is +2/-10.) - standards/persona-standards.md — §4.1 rules 4-5, the example, and the new Renames section - standards/personas/TEMPLATE/persona.yml — drop the aliases scaffold line Verification - persona.schema.json is a valid Draft 2020-12 schema; diff is +2/-10. - Conditional behaviour verified: qa-lead's post-#1289 shape (handle only) PASSES; a manifest carrying `aliases` FAILS — which is exactly why #1289 goes first. - TEMPLATE/persona.yml validates against the updated schema. - markdownlint-cli2: clean. - Zero `aliases` references remain outside the deliberate explanatory notes. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(reviews): address review comments [skip ci-relay] --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Co-authored-by: donpetry-bot <281750570+donpetry-bot@users.noreply.github.com> Co-authored-by: Don Petry Bot <donpetry+bot@gmail.com>
* feat(personas): the persona runtime — dispatch side of the mention framework Closes the loop the router opened. petry-projects/.github#754 detects an `@petry-projects/<role>` mention and fires a `persona-mention` repository_dispatch at this repo; nothing listened. This is what listens. Architecture: ONE runner, single-homed The runner is NOT fanned out to the fleet. Every mention across every repo funnels through the router's dispatch to THIS repo, so the runner lives only here and is called locally (uses: ./). That is why it needs no channel tag or canary registration — unlike the router (`persona-mention`), which IS consumed by fleet repos and rolls out through the rings. One runner serves every persona, convention-driven (prompts/<id>/advisory.md), never one workflow per persona (§4.1: per-agent workflows are the drift the framework exists to prevent). The recursion marker is closed from BOTH sides The router skips any comment carrying '<!-- persona:'. This runner's prompt and its pr_comment_has_marker guard REQUIRE every advisory's first line to be '<!-- persona:<id> -->'. So a persona's own comment can never re-summon it or any persona named in the thread — the .github-private#860 failure (1,481 acks in 4.5h), closed from the writing side. Contents - prompts/qa-lead/advisory.md — qa-lead's advisory behaviour: read the vendored bmad-tea skill by path, assess test-risk (risk vs value, no manufactured risk), post ONE advisory comment. Advisory only — no writes, no PRs, no labels. - scripts/lib/persona-runner.sh — the pure core: payload re-validation (this runner is a SEPARATE trust boundary — a repository_dispatch can be sent by anything with a token, so a malformed persona id must never reach a path interpolation), advisory-prompt resolution by convention, and the shared marker. - .github/workflows/persona-runner-reusable.yml — the runtime: resolve prompt → install claude-code → run headless → the prompt posts one marked comment. PAT (not github.token) because the item lives in a DIFFERENT repo. Rate-limit is a soft skip, matching ci-failure-analyst. - .github/workflows/persona-runner.yml — thin caller: maps the dispatch payload to the reusable's inputs, nothing else. - tests/persona_runner.bats — 15 tests, weighted to the marker and payload re-validation (path traversal, bad ids, missing prompt → rc 3 not a crash). - lint.yml — wires the suite into CI (a bats suite with no runner is unenforced). A persona with no prompts/<id>/advisory.md yet is a clean no-op ("no runtime wired"), not a failure — a persona may declare an address before its runtime exists. NOT wired live here. persona-runner-reusable.yml is workflow_call-only, so per the org standard it must be set disabled_manually after merge (GitHub ignores the flag for workflow_call). And no mention reaches it until the router release is cut and .github-private adopts the `persona-mention` stub — the `next`-ring enablement, which follows in sequence. This PR is the runtime; turning it on is the next step. Verification - bats tests/persona_runner.bats: 15/15. - shellcheck --severity=warning -x scripts/lib/persona-runner.sh: clean. - actionlint (CI's ignores) on both workflows: clean. - prompts/** is a markdownlint ignore (agent instructions, not prose). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(reviews): address review comments [skip ci-relay] * fix(personas): make the recursion marker MECHANICALLY enforced (review) Copilot's most important finding: the agent posted its OWN comment, so the marker was only prompt-REQUESTED, not enforced. An agent that forgot it would post an unmarked comment → self-summon → #860 (1,481 acks in 4.5h). My "closed from both sides" claim was false. Restructured, not patched. Read/write split - The AGENT now runs with github.token, which reads the PUBLIC source repo but CANNOT post to it (scoped to THIS repo). It PRINTS the advisory between sentinels; it never receives GH_PAT_WORKFLOWS. - The WORKFLOW extracts the body, guarantees the marker (pr_ensure_marker — prepends if the agent omitted it), and is the ONLY writer, using the PAT. So an agent that forgets the marker cannot post an unmarked comment because it cannot post at all. The guard is now mechanical. New helpers pr_extract_advisory / pr_ensure_marker, tested to the invariant "the posted body always starts with the marker". Other review fixes - Gemini (high): pr_comment_has_marker kept the trailing '\r' from GitHub's CRLF bodies, so the guard silently never matched. Strip it; CRLF test added. - Copilot: resolve step conflated rc 2 (malformed/hostile id) with rc 3 (no runtime). Now rc 2 FAILS loudly, rc 3 is the benign skip. - Copilot: concurrency group keyed only on item_number, which is empty for discussions → all discussion mentions for one persona+repo collapsed into one group. Added comment_url to the key. - Copilot: prompt used $COMMENT_URL (an html_url) with `gh api`, referenced an undefined $BODY, and put a parenthetical on the marker line. Rewritten: the agent derives the API url, prints between sentinels (no self-post, no $BODY), and the marker is alone on the first body line. - Gemini: `head -400` → `head -n 400 || true` (POSIX + no SIGPIPE under pipefail). Discussion posting is explicitly a no-op-with-notice for now (the issue-comments endpoint doesn't serve discussions); wired when the discussion path lands. Verification - bats tests/persona_runner.bats: 21/21 (15 + 6: CRLF, sentinel extraction x2, marker-prepend, unchanged-when-present, and the post-body invariant). - shellcheck --severity=warning -x scripts/lib/persona-runner.sh: clean. - actionlint (CI ignores) on both workflows: clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(reviews): address review comments [skip ci-relay] * fix(personas): restore the CRLF test; pin the COMMENT_URL contract Two regressions from the review-agent's push (b6f12fa): 1. It DELETED the CRLF test that Gemini specifically requested and that guards the '\r'-stripping fix. Restored — without it, a future revert of the strip would pass CI silently. 2. It changed the prompt to `gh api "$COMMENT_URL"` (correct target: an API url), but the router (#754, merged) currently sends `.html_url` — and `gh api` cannot fetch a github.com browser url, so the comment-body read fails. The prompt already degrades gracefully ("if that fails, proceed from the item"), so the soak still works; but the contract must be made real. The fix is in the ROUTER, not here: it should send the comment's `.url` (API) not `.html_url`. Companion PR on petry-projects/.github does that. Documented the contract in the prompt so the two sides agree. Verification - bats tests/persona_runner.bats: 22/22 (CRLF test back). - shellcheck --severity=warning -x: clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Co-authored-by: donpetry-bot <281750570+donpetry-bot@users.noreply.github.com>
Summary
Serves the addressing contract #752 defined. One router for every persona — personas do not each ship a mention workflow (§4.1). A stub per agent is exactly the drift the manifest exists to prevent, and it's how the fleet ended up with two divergent registries last time.
There is no persona index, by design
This was the open question when #752 landed, and it dissolved. Because
address.handle's slug MUST equalid, andidMUST equal the persona's directory name — both enforced byvalidate-personas.py—@petry-projects/qa-leadresolves topersonas/qa-lead/persona.ymlby convention. The manifest is the index-of-record (§1.1); nothing derives, caches, or duplicates it.It also avoids inverting the established derivation direction:
canary-rings.json(.github) →release/registry.yml(.github-private). A persona index would have had to flow the other way, putting generated content in the public standards repo.And a 404 is a meaningful answer — "not a persona". That's also how a real, non-persona team like
@petry-projects/org-leadsfalls through harmlessly instead of erroring.The manifest decides; the router restates nothing
Whether the mention surface is enabled, in what mode, behind which trust floor, under which
opt_out_label— all read from the persona. A persona may tighten the job-level trust default but never loosen it, and an undeclared floor denies, so a persona that forgot to declare trust isn't more permissive than one that did.Recursion — why the core is a tested library, not inline shell
Comments posted via a PAT re-trigger workflows, unlike
GITHUB_TOKEN..github-private#860burned 1,481 identical acks in 4.5h from a single self-loop. With N mutually addressable personas the cycles are combinatorial, not self-loops:qa-leadanswering a thread that mentionsdev-leadis enough.So the guard is two independent axes — bot actor and agent marker — enforced in both the job-level
if(so secrets never reach a bad run) andpm_should_route(so it's testable). The marker match is a prefix: #860's first fix matched one exact ack string and still self-looped through a different agent-authored comment.Contents
scripts/lib/persona-mention.shtests/persona_mention.bats.github/workflows/persona-mention-tests.ymlcanary-rollout-tests.ymlexists to close), so it ships with the suite.github/workflows/persona-mention-reusable.ymlgithub.job_workflow_sha(the #465/#528tooling_reflesson); every event field arrives viaenv, never inline${{ }}in arun:bodystandards/workflows/persona-mention.ymlpersona-mention/v1-stable(major-scoped per #657, matching live stubs)standards/canary-rings.jsonDeliberately NOT here
ring-pins.shRING_REUSABLESis untouched. Addingpersona-mentionwould makecompliance-auditdemand the stub in repos that don't have it, and make the deploy sweep report drift — manufacturing failures for a reusable with no cut release. That belongs to rollout.persona-mentionto.github-private, where nothing listens — a safe no-op.Verification (CI's exact invocations, run locally)
bats tests/persona_mention.bats— 31/31shellcheck --severity=warning -x scripts/**/*.sh— cleanactionlint -ignore 'SC2129' -ignore '"job_workflow_sha" is not defined' .github/workflows/*.yml— clean. That ignore is pre-existing and correct:job_workflow_shais a real context property actionlint 1.7.7's schema omits, and the mergedauto-rebase-reusable.ymltrips the identical false positive.markdownlint-cli2— cleancanary-rings.json— spliced as text, not re-serialized: the file's encoding is genuinely mixed (em-dashes escaped, arrows half-literal) and its agent order is insertion order, not alphabetical, so a JSON round-trip rewrote ~140 untouched lines. Result is 63 insertions / 0 deletions, structurally identical topr-review-mention's entry but forhost/reusable/run_workflow.One bug the tests caught
pm_manifest_querypiped python into jq, so the pipeline reported jq's status — a manifest parse failure returned empty output with exit 0, indistinguishable from "the manifest says no". A malformed or truncated manifest would have silently mis-routed. Now captured and checked.🤖 Generated with Claude Code
Summary by CodeRabbit
@petry-projects/<role>persona mentions across issues, PRs, reviews, and discussions.