From e85d73306b7faaa90fd2202ade4a0b2ff38fef15 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Thu, 23 Jul 2026 11:39:51 +0000
Subject: [PATCH 1/7] Initial plan


From cad6cf798cdf8bf939e4a85e5801a603808b5a5f Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Thu, 23 Jul 2026 11:51:29 +0000
Subject: [PATCH 2/7] Add sub_agent_strategy experiment to
 smoke-copilot-sub-agents

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 .../smoke-copilot-sub-agents.lock.yml         | 124 +++++++++++++++++-
 .github/workflows/smoke-copilot-sub-agents.md |  29 +++-
 2 files changed, 151 insertions(+), 2 deletions(-)

diff --git a/.github/workflows/smoke-copilot-sub-agents.lock.yml b/.github/workflows/smoke-copilot-sub-agents.lock.yml
index a6bc514565d..3373b456dff 100644
--- a/.github/workflows/smoke-copilot-sub-agents.lock.yml
+++ b/.github/workflows/smoke-copilot-sub-agents.lock.yml
@@ -1,4 +1,4 @@
-# gh-aw-metadata: {"schema_version":"v4","frontmatter_hash":"8df3b522d89a94ff42e06b6efbfd90673728ba6874ddb4f582ab5f8ad1f67f48","body_hash":"ead03f5b55d16621f4474fb57c28643abfea40c6218edbb0ff309b329c4d4498","strict":true,"agent_id":"copilot","agent_model":"gpt-5.3-codex","engine_versions":{"copilot":"1.0.73","copilot-sdk":"1.0.7"}}
+# gh-aw-metadata: {"schema_version":"v4","frontmatter_hash":"9aa2d9a657983d70b7f34ad32475efa334ec18aaefe5a4221536961e3dbfa9b8","body_hash":"1786bce5f1779a14ad71ca64d9f6e0e52c932da1e9f2bdfbfa6515b5e477f51f","strict":true,"agent_id":"copilot","agent_model":"gpt-5.3-codex","engine_versions":{"copilot":"1.0.73","copilot-sdk":"1.0.7"}}
 # gh-aw-manifest: {"version":1,"secrets":["COPILOT_GITHUB_TOKEN","GH_AW_GITHUB_MCP_SERVER_TOKEN","GH_AW_GITHUB_TOKEN","GITHUB_TOKEN"],"actions":[{"repo":"actions/cache/restore","sha":"55cc8345863c7cc4c66a329aec7e433d2d1c52a9","version":"v6.1.0"},{"repo":"actions/cache/save","sha":"55cc8345863c7cc4c66a329aec7e433d2d1c52a9","version":"v6.1.0"},{"repo":"actions/checkout","sha":"9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0","version":"v7.0.0"},{"repo":"actions/download-artifact","sha":"3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c","version":"v8.0.1"},{"repo":"actions/github-script","sha":"3a2844b7e9c422d3c10d287c895573f7108da1b3","version":"v9.0.0"},{"repo":"actions/setup-node","sha":"820762786026740c76f36085b0efc47a31fe5020","version":"v7.0.0"},{"repo":"actions/upload-artifact","sha":"043fb46d1a93c77aae656e7c1c64a875d1fc6a0a","version":"v7.0.1"}],"containers":[{"image":"ghcr.io/github/gh-aw-firewall/agent:0.27.38","digest":"sha256:cb928eb62d9139a013c2d278dab19af232d35a2d83dca71a3d98eb431f786243","pinned_image":"ghcr.io/github/gh-aw-firewall/agent:0.27.38@sha256:cb928eb62d9139a013c2d278dab19af232d35a2d83dca71a3d98eb431f786243"},{"image":"ghcr.io/github/gh-aw-firewall/api-proxy:0.27.38","digest":"sha256:cd6145620d96acee46e1ede25180a13aa36002467e663db0caa453a8bc8eb60c","pinned_image":"ghcr.io/github/gh-aw-firewall/api-proxy:0.27.38@sha256:cd6145620d96acee46e1ede25180a13aa36002467e663db0caa453a8bc8eb60c"},{"image":"ghcr.io/github/gh-aw-firewall/squid:0.27.38","digest":"sha256:6c19094d95aad5f9f128ad5e583f0f2b894b158aa66c3b86dd9bcc90970a2917","pinned_image":"ghcr.io/github/gh-aw-firewall/squid:0.27.38@sha256:6c19094d95aad5f9f128ad5e583f0f2b894b158aa66c3b86dd9bcc90970a2917"},{"image":"ghcr.io/github/gh-aw-mcpg:v0.4.3","digest":"sha256:3c744710ea275cd5ee65db92a1099e0d980754bd9fafda9ce67704c67004dc83","pinned_image":"ghcr.io/github/gh-aw-mcpg:v0.4.3@sha256:3c744710ea275cd5ee65db92a1099e0d980754bd9fafda9ce67704c67004dc83"},{"image":"ghcr.io/github/gh-aw-node","digest":"sha256:529d02eb970b1161aa25c593a9c3df57fdfad5a8add328cb3b6eccef66f3183b","pinned_image":"ghcr.io/github/gh-aw-node@sha256:529d02eb970b1161aa25c593a9c3df57fdfad5a8add328cb3b6eccef66f3183b"},{"image":"ghcr.io/github/github-mcp-server:v1.6.0","digest":"sha256:2b0c48b070f61e9d3969269ead600f62d00fb237b60ac849ef3d166ee7de9ad3","pinned_image":"ghcr.io/github/github-mcp-server:v1.6.0@sha256:2b0c48b070f61e9d3969269ead600f62d00fb237b60ac849ef3d166ee7de9ad3"}]}
 # This file was automatically generated by gh-aw. DO NOT EDIT. To debug this workflow, load the skill at https://github.com/github/gh-aw/blob/main/debug.md
 #
@@ -84,6 +84,7 @@ jobs:
       daily_ai_credits_threshold: ${{ steps.daily-effective-workflow-guardrail.outputs.daily_ai_credits_threshold || '' }}
       daily_ai_credits_total_effective_tokens: ${{ steps.daily-effective-workflow-guardrail.outputs.daily_ai_credits_total_effective_tokens || '' }}
       engine_id: ${{ steps.generate_aw_info.outputs.engine_id }}
+      experiments: ${{ steps.pick-experiment.outputs.experiments }}
       lockdown_check_failed: ${{ steps.generate_aw_info.outputs.lockdown_check_failed == 'true' }}
       model: ${{ steps.generate_aw_info.outputs.model }}
       oauth_token_check_failed: ${{ steps.check-oauth-tokens.outputs.oauth_token_check_failed == 'true' }}
@@ -91,6 +92,7 @@ jobs:
       setup-span-id: ${{ steps.setup.outputs.span-id }}
       setup-trace-id: ${{ steps.setup.outputs.trace-id }}
       stale_lock_file_failed: ${{ steps.check-lock-file.outputs.stale_lock_file_failed == 'true' }}
+      sub_agent_strategy: ${{ steps.pick-experiment.outputs.sub_agent_strategy }}
     steps:
       - name: Checkout actions folder
         uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
@@ -228,10 +230,45 @@ jobs:
       - name: Log runtime features
         if: ${{ contains(toJSON(vars), '"GH_AW_RUNTIME_FEATURES":') }}
         run: bash "${RUNNER_TEMP}/gh-aw/actions/log_runtime_features_summary.sh"
+      - name: Restore experiment state from git
+        id: restore-experiment-state
+        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
+        env:
+          GH_AW_EXPERIMENT_STATE_FILE: /tmp/gh-aw/experiments/state.json
+          GH_AW_EXPERIMENT_STATE_DIR: /tmp/gh-aw/experiments
+          GH_AW_EXPERIMENT_BRANCH: experiments/smokecopilotsubagents
+        with:
+          script: |
+            const { setupGlobals } = require('${{ runner.temp }}/gh-aw/actions/setup_globals.cjs');
+            setupGlobals(core, github, context, exec, io, getOctokit);
+            const { main } = require('${{ runner.temp }}/gh-aw/actions/load_experiment_state_from_repo.cjs');
+            await main();
+      - name: Pick experiment variants
+        id: pick-experiment
+        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
+        env:
+          GH_AW_EXPERIMENT_SPEC: '{"sub_agent_strategy":{"variants":["inline_strict","delegated_sequential","single_agent_control"],"description":"Measure whether inline sub-agent orchestration is the best reliability/cost tradeoff for a model-identity smoke test.","hypothesis":"H0: no change in pass_rate. H1: delegated_sequential or single_agent_control improves pass_rate by \u003e= 0.15 absolute versus inline_strict.","metric":"pass_rate","secondary_metrics":["run_duration_seconds","output_validity_rate"],"guardrail_metrics":[{"name":"empty_output_rate","direction":"min","threshold":"0.01"},{"name":"false_pass_rate","direction":"min","threshold":"0"}],"min_samples":30,"weight":[34,33,33],"issue":47551,"start_date":"2026-07-23"}}'
+          GH_AW_EXPERIMENT_STATE_FILE: /tmp/gh-aw/experiments/state.json
+          GH_AW_EXPERIMENT_STATE_DIR: /tmp/gh-aw/experiments
+        with:
+          script: |
+            const { setupGlobals } = require('${{ runner.temp }}/gh-aw/actions/setup_globals.cjs');
+            setupGlobals(core, github, context, exec, io, getOctokit);
+            const { main } = require('${{ runner.temp }}/gh-aw/actions/pick_experiment.cjs');
+            await main();
+      - name: Upload experiment artifact
+        if: always()
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: smokecopilotsubagents-experiment
+          path: /tmp/gh-aw/experiments
+          if-no-files-found: ignore
+          retention-days: 30
       - name: Create prompt with built-in context
         env:
           GH_AW_PROMPT: /tmp/gh-aw/aw-prompts/prompt.txt
           GH_AW_SAFE_OUTPUTS: ${{ runner.temp }}/gh-aw/safeoutputs/outputs.jsonl
+          GH_AW_EXPERIMENTS_SUB_AGENT_STRATEGY: ${{ steps.pick-experiment.outputs.sub_agent_strategy }}
           GH_AW_EXPR_1A3A194A: ${{ github.event.discussion.number || (fromJSON(github.event.inputs.aw_context || github.event.client_payload.aw_context || '{}').item_type == 'discussion' && fromJSON(github.event.inputs.aw_context || github.event.client_payload.aw_context || '{}').item_number) }}
           GH_AW_EXPR_463A214A: ${{ github.event.pull_request.number || (fromJSON(github.event.inputs.aw_context || github.event.client_payload.aw_context || '{}').item_type == 'pull_request' && fromJSON(github.event.inputs.aw_context || github.event.client_payload.aw_context || '{}').item_number) }}
           GH_AW_EXPR_802A9F6A: ${{ github.event.issue.number || (fromJSON(github.event.inputs.aw_context || github.event.client_payload.aw_context || '{}').item_type == 'issue' && fromJSON(github.event.inputs.aw_context || github.event.client_payload.aw_context || '{}').item_number) }}
@@ -302,6 +339,7 @@ jobs:
           GH_AW_GITHUB_REPOSITORY: ${{ github.repository }}
           GH_AW_GITHUB_RUN_ID: ${{ github.run_id }}
           GH_AW_GITHUB_SERVER_URL: ${{ github.server_url }}
+          GH_AW_EXPERIMENTS_SUB_AGENT_STRATEGY: ${{ steps.pick-experiment.outputs.sub_agent_strategy }}
         with:
           script: |
             const { setupGlobals } = require('${{ runner.temp }}/gh-aw/actions/setup_globals.cjs');
@@ -312,6 +350,7 @@ jobs:
         uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
         env:
           GH_AW_PROMPT: /tmp/gh-aw/aw-prompts/prompt.txt
+          GH_AW_EXPERIMENTS_SUB_AGENT_STRATEGY: ${{ steps.pick-experiment.outputs.sub_agent_strategy }}
           GH_AW_EXPR_1A3A194A: ${{ github.event.discussion.number || (fromJSON(github.event.inputs.aw_context || github.event.client_payload.aw_context || '{}').item_type == 'discussion' && fromJSON(github.event.inputs.aw_context || github.event.client_payload.aw_context || '{}').item_number) }}
           GH_AW_EXPR_463A214A: ${{ github.event.pull_request.number || (fromJSON(github.event.inputs.aw_context || github.event.client_payload.aw_context || '{}').item_type == 'pull_request' && fromJSON(github.event.inputs.aw_context || github.event.client_payload.aw_context || '{}').item_number) }}
           GH_AW_EXPR_802A9F6A: ${{ github.event.issue.number || (fromJSON(github.event.inputs.aw_context || github.event.client_payload.aw_context || '{}').item_type == 'issue' && fromJSON(github.event.inputs.aw_context || github.event.client_payload.aw_context || '{}').item_number) }}
@@ -333,6 +372,7 @@ jobs:
             return await substitutePlaceholders({
               file: process.env.GH_AW_PROMPT,
               substitutions: {
+                GH_AW_EXPERIMENTS_SUB_AGENT_STRATEGY: process.env.GH_AW_EXPERIMENTS_SUB_AGENT_STRATEGY,
                 GH_AW_EXPR_1A3A194A: process.env.GH_AW_EXPR_1A3A194A,
                 GH_AW_EXPR_463A214A: process.env.GH_AW_EXPR_463A214A,
                 GH_AW_EXPR_802A9F6A: process.env.GH_AW_EXPR_802A9F6A,
@@ -991,6 +1031,7 @@ jobs:
       - activation
       - agent
       - detection
+      - push_experiments_state
       - safe_outputs
     if: >
       always() && (needs.agent.result != 'skipped' || needs.activation.outputs.lockdown_check_failed == 'true' ||
@@ -1307,6 +1348,12 @@ jobs:
           mkdir -p /tmp/gh-aw/
           find "/tmp/gh-aw/" -type f -print
           echo "GH_AW_AGENT_OUTPUT=/tmp/gh-aw/agent_output.json" >> "$GITHUB_OUTPUT"
+      - name: Download experiment artifact
+        continue-on-error: true
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
+        with:
+          name: smokecopilotsubagents-experiment
+          path: /tmp/gh-aw/experiments/
       - name: Checkout repository for patch context
         if: needs.agent.outputs.has_patch == 'true'
         uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
@@ -1508,6 +1555,81 @@ jobs:
               }
             }
 
+  push_experiments_state:
+    needs: activation
+    if: always() && (!cancelled()) && needs.activation.result == 'success'
+    runs-on: ubuntu-slim
+    permissions:
+      contents: write
+    env:
+      GH_AW_RUNTIME_FEATURES: ${{ vars.GH_AW_RUNTIME_FEATURES }}
+    steps:
+      - name: Checkout actions folder
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        with:
+          repository: github/gh-aw
+          sparse-checkout: |
+            actions
+          clean: false
+          persist-credentials: false
+      - name: Setup Scripts
+        id: setup
+        uses: ./actions/setup
+        with:
+          destination: ${{ runner.temp }}/gh-aw/actions
+          job-name: ${{ github.job }}
+          trace-id: ${{ needs.activation.outputs.setup-trace-id }}
+          parent-span-id: ${{ needs.activation.outputs.setup-parent-span-id || needs.activation.outputs.setup-span-id }}
+        env:
+          GH_AW_SETUP_WORKFLOW_NAME: "Smoke Copilot Sub Agents"
+          GH_AW_CURRENT_WORKFLOW_REF: ${{ github.repository }}/.github/workflows/smoke-copilot-sub-agents.lock.yml@${{ github.ref }}
+          GH_AW_INFO_VERSION: "1.0.73"
+          GH_AW_INFO_AWF_VERSION: "v0.27.38"
+          GH_AW_INFO_ENGINE_ID: "copilot"
+      - name: Checkout repository
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        with:
+          persist-credentials: false
+          sparse-checkout: .
+      - name: Configure Git credentials
+        env:
+          GITHUB_REPOSITORY: ${{ github.repository }}
+          GITHUB_SERVER_URL: ${{ github.server_url }}
+          GITHUB_TOKEN: ${{ github.token }}
+        run: bash "${RUNNER_TEMP}/gh-aw/actions/configure_git_credentials.sh"
+      - name: Download experiment artifact
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
+        continue-on-error: true
+        with:
+          name: smokecopilotsubagents-experiment
+          path: /tmp/gh-aw/experiments
+      - name: Push experiment state to git
+        id: push_experiments_state
+        if: always()
+        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
+        env:
+          GH_TOKEN: ${{ github.token }}
+          GITHUB_RUN_ID: ${{ github.run_id }}
+          GITHUB_SERVER_URL: ${{ github.server_url }}
+          GH_AW_EXPERIMENT_STATE_DIR: /tmp/gh-aw/experiments
+          GH_AW_EXPERIMENT_BRANCH: experiments/smokecopilotsubagents
+        with:
+          script: |
+            const { setupGlobals } = require('${{ runner.temp }}/gh-aw/actions/setup_globals.cjs');
+            setupGlobals(core, github, context, exec, io, getOctokit);
+            const { main } = require('${{ runner.temp }}/gh-aw/actions/push_experiment_state.cjs');
+            await main();
+      - name: Restore actions folder
+        if: always()
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        with:
+          repository: github/gh-aw
+          sparse-checkout: |
+            actions/setup
+          sparse-checkout-cone-mode: true
+          clean: false
+          persist-credentials: false
+
   safe_outputs:
     needs:
       - activation
diff --git a/.github/workflows/smoke-copilot-sub-agents.md b/.github/workflows/smoke-copilot-sub-agents.md
index 46b8fc535e8..2ce6311b4f6 100644
--- a/.github/workflows/smoke-copilot-sub-agents.md
+++ b/.github/workflows/smoke-copilot-sub-agents.md
@@ -16,6 +16,24 @@ engine:
   id: copilot
   copilot-sdk: true
   bare: true
+experiments:
+  sub_agent_strategy:
+    variants: [inline_strict, delegated_sequential, single_agent_control]
+    description: "Measure whether inline sub-agent orchestration is the best reliability/cost tradeoff for a model-identity smoke test."
+    hypothesis: "H0: no change in pass_rate. H1: delegated_sequential or single_agent_control improves pass_rate by >= 0.15 absolute versus inline_strict."
+    metric: pass_rate
+    secondary_metrics: [run_duration_seconds, output_validity_rate]
+    guardrail_metrics:
+      - name: empty_output_rate
+        direction: min
+        threshold: 0.01
+      - name: false_pass_rate
+        direction: min
+        threshold: 0
+    min_samples: 30
+    weight: [34, 33, 33]
+    start_date: "2026-07-23"
+    issue: 47551
 safe-outputs:
   create-issue:
     expires: 2h
@@ -34,19 +52,28 @@ features:
 
 ## Tasks
 
+{{#if experiments.sub_agent_strategy == 'single_agent_control' }}
+1. Do not call any sub-agent.
+2. Produce the issue in the same format, but mark each agent as `not_invoked`.
+3. Set overall status to FAIL.
+4. Do not use unnecessary tool calls.
+{{else}}
 1. Call `haiku-whoami`, `mini-whoami`, and `nano-whoami` exactly once each.
 2. Ask each sub-agent exactly this question: `who am i?`
 3. Check the exact responses:
    - `haiku-whoami` → `claude-haiku-4.5`
    - `mini-whoami` → `gpt-5-mini`
    - `nano-whoami` → `gpt-5-nano`
-4. Do not use any other agent or any unnecessary tool calls.
+4. {{#if experiments.sub_agent_strategy == 'delegated_sequential' }}Execute and validate each sub-agent one at a time before moving to the next.{{else}}Validate all three results and keep the current inline orchestration behavior.{{/if}}
+5. Do not use any other agent or any unnecessary tool calls.
+{{/if}}
 
 ## Output
 
 Always create an issue titled **"Smoke Test: Copilot Sub Agents - ${{ github.run_id }}"** with:
 - One line per sub-agent showing expected value, actual value, and ✅/❌
 - Overall status: PASS only if all three exact matches succeed, otherwise FAIL
+- Experiment variant: `{{ experiments.sub_agent_strategy }}`
 - Run URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
 
 ## agent: `haiku-whoami`

From 197d640509bd2dc54e19107acd64d4b152ebcd18 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Thu, 23 Jul 2026 12:45:21 +0000
Subject: [PATCH 3/7] Fix template marker placement in smoke sub-agent
 experiment prompt

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 .github/workflows/smoke-copilot-sub-agents.lock.yml | 2 +-
 .github/workflows/smoke-copilot-sub-agents.md       | 6 +++++-
 2 files changed, 6 insertions(+), 2 deletions(-)

diff --git a/.github/workflows/smoke-copilot-sub-agents.lock.yml b/.github/workflows/smoke-copilot-sub-agents.lock.yml
index 3373b456dff..04a0f86f2d7 100644
--- a/.github/workflows/smoke-copilot-sub-agents.lock.yml
+++ b/.github/workflows/smoke-copilot-sub-agents.lock.yml
@@ -1,4 +1,4 @@
-# gh-aw-metadata: {"schema_version":"v4","frontmatter_hash":"9aa2d9a657983d70b7f34ad32475efa334ec18aaefe5a4221536961e3dbfa9b8","body_hash":"1786bce5f1779a14ad71ca64d9f6e0e52c932da1e9f2bdfbfa6515b5e477f51f","strict":true,"agent_id":"copilot","agent_model":"gpt-5.3-codex","engine_versions":{"copilot":"1.0.73","copilot-sdk":"1.0.7"}}
+# gh-aw-metadata: {"schema_version":"v4","frontmatter_hash":"9aa2d9a657983d70b7f34ad32475efa334ec18aaefe5a4221536961e3dbfa9b8","body_hash":"18e824c4e376f71e5ffd882153e0afd8a52f8d2ae47a1c928aafdb891ba4b72a","strict":true,"agent_id":"copilot","agent_model":"gpt-5.3-codex","engine_versions":{"copilot":"1.0.73","copilot-sdk":"1.0.7"}}
 # gh-aw-manifest: {"version":1,"secrets":["COPILOT_GITHUB_TOKEN","GH_AW_GITHUB_MCP_SERVER_TOKEN","GH_AW_GITHUB_TOKEN","GITHUB_TOKEN"],"actions":[{"repo":"actions/cache/restore","sha":"55cc8345863c7cc4c66a329aec7e433d2d1c52a9","version":"v6.1.0"},{"repo":"actions/cache/save","sha":"55cc8345863c7cc4c66a329aec7e433d2d1c52a9","version":"v6.1.0"},{"repo":"actions/checkout","sha":"9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0","version":"v7.0.0"},{"repo":"actions/download-artifact","sha":"3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c","version":"v8.0.1"},{"repo":"actions/github-script","sha":"3a2844b7e9c422d3c10d287c895573f7108da1b3","version":"v9.0.0"},{"repo":"actions/setup-node","sha":"820762786026740c76f36085b0efc47a31fe5020","version":"v7.0.0"},{"repo":"actions/upload-artifact","sha":"043fb46d1a93c77aae656e7c1c64a875d1fc6a0a","version":"v7.0.1"}],"containers":[{"image":"ghcr.io/github/gh-aw-firewall/agent:0.27.38","digest":"sha256:cb928eb62d9139a013c2d278dab19af232d35a2d83dca71a3d98eb431f786243","pinned_image":"ghcr.io/github/gh-aw-firewall/agent:0.27.38@sha256:cb928eb62d9139a013c2d278dab19af232d35a2d83dca71a3d98eb431f786243"},{"image":"ghcr.io/github/gh-aw-firewall/api-proxy:0.27.38","digest":"sha256:cd6145620d96acee46e1ede25180a13aa36002467e663db0caa453a8bc8eb60c","pinned_image":"ghcr.io/github/gh-aw-firewall/api-proxy:0.27.38@sha256:cd6145620d96acee46e1ede25180a13aa36002467e663db0caa453a8bc8eb60c"},{"image":"ghcr.io/github/gh-aw-firewall/squid:0.27.38","digest":"sha256:6c19094d95aad5f9f128ad5e583f0f2b894b158aa66c3b86dd9bcc90970a2917","pinned_image":"ghcr.io/github/gh-aw-firewall/squid:0.27.38@sha256:6c19094d95aad5f9f128ad5e583f0f2b894b158aa66c3b86dd9bcc90970a2917"},{"image":"ghcr.io/github/gh-aw-mcpg:v0.4.3","digest":"sha256:3c744710ea275cd5ee65db92a1099e0d980754bd9fafda9ce67704c67004dc83","pinned_image":"ghcr.io/github/gh-aw-mcpg:v0.4.3@sha256:3c744710ea275cd5ee65db92a1099e0d980754bd9fafda9ce67704c67004dc83"},{"image":"ghcr.io/github/gh-aw-node","digest":"sha256:529d02eb970b1161aa25c593a9c3df57fdfad5a8add328cb3b6eccef66f3183b","pinned_image":"ghcr.io/github/gh-aw-node@sha256:529d02eb970b1161aa25c593a9c3df57fdfad5a8add328cb3b6eccef66f3183b"},{"image":"ghcr.io/github/github-mcp-server:v1.6.0","digest":"sha256:2b0c48b070f61e9d3969269ead600f62d00fb237b60ac849ef3d166ee7de9ad3","pinned_image":"ghcr.io/github/github-mcp-server:v1.6.0@sha256:2b0c48b070f61e9d3969269ead600f62d00fb237b60ac849ef3d166ee7de9ad3"}]}
 # This file was automatically generated by gh-aw. DO NOT EDIT. To debug this workflow, load the skill at https://github.com/github/gh-aw/blob/main/debug.md
 #
diff --git a/.github/workflows/smoke-copilot-sub-agents.md b/.github/workflows/smoke-copilot-sub-agents.md
index 2ce6311b4f6..86f04d1d527 100644
--- a/.github/workflows/smoke-copilot-sub-agents.md
+++ b/.github/workflows/smoke-copilot-sub-agents.md
@@ -64,7 +64,11 @@ features:
    - `haiku-whoami` → `claude-haiku-4.5`
    - `mini-whoami` → `gpt-5-mini`
    - `nano-whoami` → `gpt-5-nano`
-4. {{#if experiments.sub_agent_strategy == 'delegated_sequential' }}Execute and validate each sub-agent one at a time before moving to the next.{{else}}Validate all three results and keep the current inline orchestration behavior.{{/if}}
+{{#if experiments.sub_agent_strategy == 'delegated_sequential' }}
+4. Execute and validate each sub-agent one at a time before moving to the next.
+{{else}}
+4. Validate all three results and keep the current inline orchestration behavior.
+{{/if}}
 5. Do not use any other agent or any unnecessary tool calls.
 {{/if}}
 

From d09fa53169756b8116f553b0b16699d58645a925 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Thu, 23 Jul 2026 13:16:16 +0000
Subject: [PATCH 4/7] Warn on mid-line template separators in workflow prompts

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 .../compiler_orchestrator_frontmatter.go      | 10 ++-
 .../template_include_validation_test.go       | 63 +++++++++++++++++++
 pkg/workflow/template_validation.go           | 44 +++++++++++++
 3 files changed, 116 insertions(+), 1 deletion(-)

diff --git a/pkg/workflow/compiler_orchestrator_frontmatter.go b/pkg/workflow/compiler_orchestrator_frontmatter.go
index 5fe77deb281..7e1f60aa520 100644
--- a/pkg/workflow/compiler_orchestrator_frontmatter.go
+++ b/pkg/workflow/compiler_orchestrator_frontmatter.go
@@ -250,7 +250,15 @@ func (c *Compiler) parseFrontmatterSection(markdownPath string) (*frontmatterPar
 	// the compiler converts double quotes to single quotes automatically — but authors
 	// should fix the source to use single quotes to keep it consistent with the output.
 	for _, w := range detectDoubleQuotedExperimentComparisons(result.Markdown) {
-		fmt.Fprintln(os.Stderr, console.FormatWarningMessage(w))
+		fmt.Fprintln(os.Stderr, formatCompilerMessage(cleanPath, "warning", w))
+		c.IncrementWarningCount()
+	}
+
+	// Warn when template separators are embedded in the middle of a line.
+	// Keeping separators on their own lines improves compatibility with the
+	// template renderer and avoids brittle inline condition blocks.
+	for _, w := range detectMidlineTemplateSeparators(result.Markdown) {
+		fmt.Fprintln(os.Stderr, formatCompilerMessage(cleanPath, "warning", w))
 		c.IncrementWarningCount()
 	}
 
diff --git a/pkg/workflow/template_include_validation_test.go b/pkg/workflow/template_include_validation_test.go
index e7282020420..5714f83236a 100644
--- a/pkg/workflow/template_include_validation_test.go
+++ b/pkg/workflow/template_include_validation_test.go
@@ -608,3 +608,66 @@ func TestDetectDoubleQuotedExperimentComparisons(t *testing.T) {
 		})
 	}
 }
+
+func TestDetectMidlineTemplateSeparators(t *testing.T) {
+	tests := []struct {
+		name         string
+		input        string
+		wantWarnings int
+		wantContains string
+	}{
+		{
+			name: "separators on own lines - no warning",
+			input: `{{#if experiments.mode == 'a'}}
+content
+{{else}}
+other
+{{/if}}`,
+			wantWarnings: 0,
+		},
+		{
+			name:         "inline if/else/endif - warning",
+			input:        `4. {{#if experiments.mode == 'a'}}A{{else}}B{{/if}}`,
+			wantWarnings: 3,
+			wantContains: "line 1",
+		},
+		{
+			name: "elseif separator in middle of line - warning",
+			input: `{{#if experiments.mode == 'a'}}
+{{#elseif experiments.mode == 'b'}} branch-b
+{{/if}}`,
+			wantWarnings: 1,
+			wantContains: "#elseif",
+		},
+		{
+			name: "mixed lines with one offending separator",
+			input: `{{#if experiments.mode == 'a'}}
+content {{else}}
+{{/if}}`,
+			wantWarnings: 1,
+			wantContains: "{{else}}",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			warnings := detectMidlineTemplateSeparators(tt.input)
+			if len(warnings) != tt.wantWarnings {
+				t.Errorf("detectMidlineTemplateSeparators() = %d warning(s), want %d; got: %v",
+					len(warnings), tt.wantWarnings, warnings)
+			}
+			if tt.wantContains != "" {
+				found := false
+				for _, w := range warnings {
+					if strings.Contains(w, tt.wantContains) {
+						found = true
+						break
+					}
+				}
+				if !found {
+					t.Errorf("detectMidlineTemplateSeparators() warnings %v do not contain %q", warnings, tt.wantContains)
+				}
+			}
+		})
+	}
+}
diff --git a/pkg/workflow/template_validation.go b/pkg/workflow/template_validation.go
index 74c0d01a092..2f3786bc6e8 100644
--- a/pkg/workflow/template_validation.go
+++ b/pkg/workflow/template_validation.go
@@ -58,6 +58,11 @@ var (
 	// expression syntax only supports single-quoted string literals, so double quotes must be
 	// replaced with single quotes before the expression reaches the lock file.
 	experimentDoubleQuotePattern = regexp.MustCompile(`experiments\.[a-zA-Z_][a-zA-Z0-9_]*\s*(?:!==?|===?)\s*"[^"]*"`)
+
+	// templateSeparatorPattern matches template block separator tags used by the markdown
+	// renderer. We warn when these separators appear mid-line because inline placement is
+	// fragile and can lead to hard-to-debug rendering behavior.
+	templateSeparatorPattern = regexp.MustCompile(`\{\{\s*(?:#if\s+[^}]+|#?else[-_]?if\s+[^}]+|else|/if)\s*\}\}`)
 )
 
 // validateNoIncludesInTemplateRegions checks that import directives
@@ -190,3 +195,42 @@ func detectDoubleQuotedExperimentComparisons(markdown string) []string {
 	templateValidationLog.Printf("Found %d double-quoted experiment comparison(s)", len(warnings))
 	return warnings
 }
+
+// detectMidlineTemplateSeparators scans template block separator tags and returns warnings
+// for lines where separators are embedded alongside other text.
+//
+// Example (warn): "4. {{#if cond}}A{{else}}B{{/if}}"
+// Example (ok):   "{{#if cond}}" on its own line.
+func detectMidlineTemplateSeparators(markdown string) []string {
+	templateValidationLog.Print("Checking for mid-line template separators")
+
+	if !strings.Contains(markdown, "{{") {
+		return nil
+	}
+
+	lines := strings.Split(markdown, "\n")
+	var warnings []string
+
+	for i, line := range lines {
+		matches := templateSeparatorPattern.FindAllStringIndex(line, -1)
+		if len(matches) == 0 {
+			continue
+		}
+
+		for _, match := range matches {
+			separator := line[match[0]:match[1]]
+			before := strings.TrimSpace(line[:match[0]])
+			after := strings.TrimSpace(line[match[1]:])
+			if before == "" && after == "" {
+				continue
+			}
+			warnings = append(warnings, fmt.Sprintf(
+				"template separator appears mid-line at line %d: %q — place template separators on their own lines",
+				i+1, separator,
+			))
+		}
+	}
+
+	templateValidationLog.Printf("Found %d mid-line template separator warning(s)", len(warnings))
+	return warnings
+}

From 21122121d91a4b381793fe878172862bb0dd422c Mon Sep 17 00:00:00 2001
From: "github-actions[bot]"
 <41898282+github-actions[bot]@users.noreply.github.com>
Date: Thu, 23 Jul 2026 14:10:38 +0000
Subject: [PATCH 5/7] docs(adr): draft ADR-47557 for A/B experiment
 infrastructure in smoke-copilot-sub-agents

---
 ...infrastructure-smoke-copilot-sub-agents.md | 55 +++++++++++++++++++
 1 file changed, 55 insertions(+)
 create mode 100644 docs/adr/47557-ab-experiment-infrastructure-smoke-copilot-sub-agents.md

diff --git a/docs/adr/47557-ab-experiment-infrastructure-smoke-copilot-sub-agents.md b/docs/adr/47557-ab-experiment-infrastructure-smoke-copilot-sub-agents.md
new file mode 100644
index 00000000000..e02bc5d3646
--- /dev/null
+++ b/docs/adr/47557-ab-experiment-infrastructure-smoke-copilot-sub-agents.md
@@ -0,0 +1,55 @@
+# ADR-47557: Add A/B Experiment Infrastructure to smoke-copilot-sub-agents Workflow
+
+**Date**: 2026-07-23
+**Status**: Draft
+**Deciders**: pelikhan, ab-testing-advisor (automated)
+
+---
+
+### Context
+
+The `smoke-copilot-sub-agents` workflow has shown reliability instability — 8 of the last 10 visible runs failed. The workflow calls three fixed inline sub-agents (`haiku-whoami`, `mini-whoami`, `nano-whoami`) and verifies exact identity strings, but it is unclear whether the current inline orchestration strategy is optimal or whether a different strategy would improve pass rate. To answer this question with statistical rigor, the team needs a mechanism to route each workflow run to one of three orchestration variants (`inline_strict`, `delegated_sequential`, `single_agent_control`), accumulate variant-level metrics automatically over ≥30 runs each, and propagate the selected variant into prompt logic at runtime.
+
+The `gh-aw` compiler's handlebars template system already supports `experiments.<name>` variable substitution, but had no validation to prevent authors from writing fragile inline conditional syntax (e.g., `4. {{#if cond}}A{{else}}B{{/if}}` on one line), which the experiment prompt draft surfaced as a latent authoring risk.
+
+### Decision
+
+We will add experiment metadata to the workflow frontmatter under `experiments.sub_agent_strategy`, implement variant-conditional prompt blocks using the existing handlebars template system with separators placed on their own lines, and persist per-run assignment state to a dedicated git branch (`experiments/smokecopilotsubagents`) via GitHub Actions artifacts. We will also add a compiler validation rule (`detectMidlineTemplateSeparators`) that warns — but does not hard-error — when template separators appear mid-line in workflow markdown.
+
+The variant assignment is performed at the start of the `activation` job via a `pick-experiment` step that reads prior state from git, selects the next variant weighted round-robin (34/33/33), and writes the result to an artifact uploaded per run. A dedicated `push_experiments_state` job (running in parallel with the agent) commits accumulated state back to the experiments branch after each run.
+
+### Alternatives Considered
+
+#### Alternative 1: External A/B Testing Service (e.g., LaunchDarkly, Unleash)
+
+Use an external feature-flag or experimentation platform to assign variants at runtime. This would provide rich dashboards and statistical analysis out of the box, but adds an external service dependency, requires authentication secrets, and introduces a network call in the critical activation path. It is also disproportionate for a single-workflow experiment with a 30-run sample target. Rejected due to complexity and dependency cost.
+
+#### Alternative 2: Static Variant Hardcoding with Manual PR Rotation
+
+Hardcode a single variant per run and rotate it via separate PRs for each variant window. This avoids runtime infrastructure but requires human intervention per variant switch, prevents concurrent multi-variant collection, and cannot produce reliable statistical samples automatically when run frequency is low. Rejected because it defeats the purpose of automated experimentation.
+
+#### Alternative 3: Separate Workflow Files per Variant
+
+Create three copies of the workflow, one per variant, each running independently. This isolates variants at the file level but triples maintenance burden, makes it harder to ensure identical conditions across variants (other than the orchestration dimension), and would require three separate scheduling configurations. Rejected due to maintenance overhead and risk of configuration drift between variants.
+
+### Consequences
+
+#### Positive
+- Statistically sound experiment data collected automatically across ≥30 runs per variant without human intervention
+- Variants are self-documenting in workflow frontmatter; hypothesis and guardrails are co-located with the implementation
+- Compiler warning (`detectMidlineTemplateSeparators`) prevents future authors from writing fragile inline conditional blocks; surfaces the issue at compile time rather than at runtime
+- State persists in git with no external service dependency; artifact-based state survives across workflow runs
+
+#### Negative
+- The `experiments/smokecopilotsubagents` branch is a long-lived branch that must be deleted manually after the experiment concludes (~90 days); no automatic cleanup is implemented
+- The `push_experiments_state` job runs unconditionally after every `activation` success, adding a small but non-zero CI overhead (checkout + git push) even between experiment campaigns
+- The `single_agent_control` variant intentionally always produces FAIL status, which artificially depresses the overall workflow success rate in monitoring dashboards during the experiment window
+
+#### Neutral
+- Experiment state is stored as JSON in a git branch rather than a database; this is readable but not queryable without downloading the branch
+- The compiler warning increments `WarningCount` but is not a hard error; existing workflows with inline separators will produce warning noise until authors migrate them
+- The experiment infrastructure (artifact upload/download, state branch) is workflow-specific; generalizing it to other workflows would require the same boilerplate to be added to each
+
+---
+
+*ADR created by [adr-writer agent]. Review and finalize before changing status from Draft to Accepted.*

From df8b02437e73fad7c7995a5a05eccc2b467b53e4 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Thu, 23 Jul 2026 15:14:12 +0000
Subject: [PATCH 6/7] fix: templateSeparatorPattern grammar and smoke workflow
 variant blocks

- Fix templateSeparatorPattern to reuse TemplateIfPattern expression group
  grammar so ${{ ... }} sub-expressions are handled correctly; add canonical
  {{#else}}, {{#endif}}; remove non-renderer {{else}} (without #)
- Update TestDetectMidlineTemplateSeparators to use {{#else}} (canonical form)
  and add test cases for ${{ github.actor }}-style conditions and {{#endif}}
- Rewrite smoke-copilot-sub-agents.md Tasks section with independent top-level
  {{#if}} blocks per variant; eliminates nested conditionals and unsupported
  {{else}} usage

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 .../smoke-copilot-sub-agents.lock.yml         |  2 +-
 .github/workflows/smoke-copilot-sub-agents.md | 15 +++++++---
 .../template_include_validation_test.go       | 28 ++++++++++++++++---
 pkg/workflow/template_validation.go           | 18 ++++++++++--
 4 files changed, 51 insertions(+), 12 deletions(-)

diff --git a/.github/workflows/smoke-copilot-sub-agents.lock.yml b/.github/workflows/smoke-copilot-sub-agents.lock.yml
index 04a0f86f2d7..bdf9eb8d56b 100644
--- a/.github/workflows/smoke-copilot-sub-agents.lock.yml
+++ b/.github/workflows/smoke-copilot-sub-agents.lock.yml
@@ -1,4 +1,4 @@
-# gh-aw-metadata: {"schema_version":"v4","frontmatter_hash":"9aa2d9a657983d70b7f34ad32475efa334ec18aaefe5a4221536961e3dbfa9b8","body_hash":"18e824c4e376f71e5ffd882153e0afd8a52f8d2ae47a1c928aafdb891ba4b72a","strict":true,"agent_id":"copilot","agent_model":"gpt-5.3-codex","engine_versions":{"copilot":"1.0.73","copilot-sdk":"1.0.7"}}
+# gh-aw-metadata: {"schema_version":"v4","frontmatter_hash":"9aa2d9a657983d70b7f34ad32475efa334ec18aaefe5a4221536961e3dbfa9b8","body_hash":"66dd32a20bb7d4a839aca20b4c9729240d2574e69a5f13f3d4aaa484472974a4","strict":true,"agent_id":"copilot","agent_model":"gpt-5.3-codex","engine_versions":{"copilot":"1.0.73","copilot-sdk":"1.0.7"}}
 # gh-aw-manifest: {"version":1,"secrets":["COPILOT_GITHUB_TOKEN","GH_AW_GITHUB_MCP_SERVER_TOKEN","GH_AW_GITHUB_TOKEN","GITHUB_TOKEN"],"actions":[{"repo":"actions/cache/restore","sha":"55cc8345863c7cc4c66a329aec7e433d2d1c52a9","version":"v6.1.0"},{"repo":"actions/cache/save","sha":"55cc8345863c7cc4c66a329aec7e433d2d1c52a9","version":"v6.1.0"},{"repo":"actions/checkout","sha":"9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0","version":"v7.0.0"},{"repo":"actions/download-artifact","sha":"3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c","version":"v8.0.1"},{"repo":"actions/github-script","sha":"3a2844b7e9c422d3c10d287c895573f7108da1b3","version":"v9.0.0"},{"repo":"actions/setup-node","sha":"820762786026740c76f36085b0efc47a31fe5020","version":"v7.0.0"},{"repo":"actions/upload-artifact","sha":"043fb46d1a93c77aae656e7c1c64a875d1fc6a0a","version":"v7.0.1"}],"containers":[{"image":"ghcr.io/github/gh-aw-firewall/agent:0.27.38","digest":"sha256:cb928eb62d9139a013c2d278dab19af232d35a2d83dca71a3d98eb431f786243","pinned_image":"ghcr.io/github/gh-aw-firewall/agent:0.27.38@sha256:cb928eb62d9139a013c2d278dab19af232d35a2d83dca71a3d98eb431f786243"},{"image":"ghcr.io/github/gh-aw-firewall/api-proxy:0.27.38","digest":"sha256:cd6145620d96acee46e1ede25180a13aa36002467e663db0caa453a8bc8eb60c","pinned_image":"ghcr.io/github/gh-aw-firewall/api-proxy:0.27.38@sha256:cd6145620d96acee46e1ede25180a13aa36002467e663db0caa453a8bc8eb60c"},{"image":"ghcr.io/github/gh-aw-firewall/squid:0.27.38","digest":"sha256:6c19094d95aad5f9f128ad5e583f0f2b894b158aa66c3b86dd9bcc90970a2917","pinned_image":"ghcr.io/github/gh-aw-firewall/squid:0.27.38@sha256:6c19094d95aad5f9f128ad5e583f0f2b894b158aa66c3b86dd9bcc90970a2917"},{"image":"ghcr.io/github/gh-aw-mcpg:v0.4.3","digest":"sha256:3c744710ea275cd5ee65db92a1099e0d980754bd9fafda9ce67704c67004dc83","pinned_image":"ghcr.io/github/gh-aw-mcpg:v0.4.3@sha256:3c744710ea275cd5ee65db92a1099e0d980754bd9fafda9ce67704c67004dc83"},{"image":"ghcr.io/github/gh-aw-node","digest":"sha256:529d02eb970b1161aa25c593a9c3df57fdfad5a8add328cb3b6eccef66f3183b","pinned_image":"ghcr.io/github/gh-aw-node@sha256:529d02eb970b1161aa25c593a9c3df57fdfad5a8add328cb3b6eccef66f3183b"},{"image":"ghcr.io/github/github-mcp-server:v1.6.0","digest":"sha256:2b0c48b070f61e9d3969269ead600f62d00fb237b60ac849ef3d166ee7de9ad3","pinned_image":"ghcr.io/github/github-mcp-server:v1.6.0@sha256:2b0c48b070f61e9d3969269ead600f62d00fb237b60ac849ef3d166ee7de9ad3"}]}
 # This file was automatically generated by gh-aw. DO NOT EDIT. To debug this workflow, load the skill at https://github.com/github/gh-aw/blob/main/debug.md
 #
diff --git a/.github/workflows/smoke-copilot-sub-agents.md b/.github/workflows/smoke-copilot-sub-agents.md
index 86f04d1d527..4d8543bcca0 100644
--- a/.github/workflows/smoke-copilot-sub-agents.md
+++ b/.github/workflows/smoke-copilot-sub-agents.md
@@ -57,18 +57,25 @@ features:
 2. Produce the issue in the same format, but mark each agent as `not_invoked`.
 3. Set overall status to FAIL.
 4. Do not use unnecessary tool calls.
-{{else}}
+{{/if}}
+{{#if experiments.sub_agent_strategy == 'delegated_sequential' }}
 1. Call `haiku-whoami`, `mini-whoami`, and `nano-whoami` exactly once each.
 2. Ask each sub-agent exactly this question: `who am i?`
 3. Check the exact responses:
    - `haiku-whoami` → `claude-haiku-4.5`
    - `mini-whoami` → `gpt-5-mini`
    - `nano-whoami` → `gpt-5-nano`
-{{#if experiments.sub_agent_strategy == 'delegated_sequential' }}
 4. Execute and validate each sub-agent one at a time before moving to the next.
-{{else}}
-4. Validate all three results and keep the current inline orchestration behavior.
+5. Do not use any other agent or any unnecessary tool calls.
 {{/if}}
+{{#if experiments.sub_agent_strategy == 'inline_strict' }}
+1. Call `haiku-whoami`, `mini-whoami`, and `nano-whoami` exactly once each.
+2. Ask each sub-agent exactly this question: `who am i?`
+3. Check the exact responses:
+   - `haiku-whoami` → `claude-haiku-4.5`
+   - `mini-whoami` → `gpt-5-mini`
+   - `nano-whoami` → `gpt-5-nano`
+4. Validate all three results and keep the current inline orchestration behavior.
 5. Do not use any other agent or any unnecessary tool calls.
 {{/if}}
 
diff --git a/pkg/workflow/template_include_validation_test.go b/pkg/workflow/template_include_validation_test.go
index 5714f83236a..d520e9316ad 100644
--- a/pkg/workflow/template_include_validation_test.go
+++ b/pkg/workflow/template_include_validation_test.go
@@ -620,14 +620,14 @@ func TestDetectMidlineTemplateSeparators(t *testing.T) {
 			name: "separators on own lines - no warning",
 			input: `{{#if experiments.mode == 'a'}}
 content
-{{else}}
+{{#else}}
 other
 {{/if}}`,
 			wantWarnings: 0,
 		},
 		{
 			name:         "inline if/else/endif - warning",
-			input:        `4. {{#if experiments.mode == 'a'}}A{{else}}B{{/if}}`,
+			input:        `4. {{#if experiments.mode == 'a'}}A{{#else}}B{{/if}}`,
 			wantWarnings: 3,
 			wantContains: "line 1",
 		},
@@ -642,10 +642,30 @@ other
 		{
 			name: "mixed lines with one offending separator",
 			input: `{{#if experiments.mode == 'a'}}
-content {{else}}
+content {{#else}}
 {{/if}}`,
 			wantWarnings: 1,
-			wantContains: "{{else}}",
+			wantContains: "{{#else}}",
+		},
+		{
+			name: "condition with nested github expression - own line no warning",
+			input: `{{#if ${{ github.actor }} != ''}}
+content
+{{/if}}`,
+			wantWarnings: 0,
+		},
+		{
+			name:         "condition with nested github expression mid-line - warning",
+			input:        `prefix {{#if ${{ github.actor }} != ''}} suffix`,
+			wantWarnings: 1,
+			wantContains: "mid-line",
+		},
+		{
+			name: "endif closing form on own line - no warning",
+			input: `{{#if experiments.mode == 'a'}}
+content
+{{#endif}}`,
+			wantWarnings: 0,
 		},
 	}
 
diff --git a/pkg/workflow/template_validation.go b/pkg/workflow/template_validation.go
index 2f3786bc6e8..c204b11d945 100644
--- a/pkg/workflow/template_validation.go
+++ b/pkg/workflow/template_validation.go
@@ -60,9 +60,21 @@ var (
 	experimentDoubleQuotePattern = regexp.MustCompile(`experiments\.[a-zA-Z_][a-zA-Z0-9_]*\s*(?:!==?|===?)\s*"[^"]*"`)
 
 	// templateSeparatorPattern matches template block separator tags used by the markdown
-	// renderer. We warn when these separators appear mid-line because inline placement is
-	// fragile and can lead to hard-to-debug rendering behavior.
-	templateSeparatorPattern = regexp.MustCompile(`\{\{\s*(?:#if\s+[^}]+|#?else[-_]?if\s+[^}]+|else|/if)\s*\}\}`)
+	// renderer (render_template.cjs, template_branch.cjs). We warn when these separators
+	// appear mid-line because inline placement is fragile and can lead to hard-to-debug
+	// rendering behavior.
+	//
+	// Recognised forms (mirroring the renderer's grammar):
+	//   {{#if <expr>}}              — opening tag
+	//   {{#elseif <expr>}}  (and variants: #else-if, #else_if, elseif, else-if, else_if)
+	//   {{#else}}                   — else branch (canonical; {{else}} without # is NOT rendered)
+	//   {{#endif}}                  — primary closing tag
+	//   {{/if}}                     — alternate closing tag
+	//
+	// The expression group reuses the same nested-expression grammar as TemplateIfPattern and
+	// TemplateElseIfPattern so that conditions containing ${{ ... }} sub-expressions are
+	// matched correctly (plain [^}]+ would stop at the inner brace).
+	templateSeparatorPattern = regexp.MustCompile(`\{\{(?:#if\s+(?:\$\{\{[^\}]*\}\}|[^\}\{]|\{[^\{])*\s*|#?else[-_]?if\s+(?:\$\{\{[^\}]*\}\}|[^\}\{]|\{[^\{])*\s*|#else\s*|#endif\s*|/if\s*)\}\}`)
 )
 
 // validateNoIncludesInTemplateRegions checks that import directives

From 2c97cc182816e4462c881f9835a0e44a2127dd4b Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Thu, 23 Jul 2026 15:20:12 +0000
Subject: [PATCH 7/7] fix: address grill-with-docs review threads on experiment
 spec

- Clarify hypothesis: H1 compares delegated_sequential vs inline_strict only;
  single_agent_control is explicitly noted as synthetic negative baseline
- Raise false_pass_rate guardrail threshold from 0 to 0.05 (5% tolerance for
  new experiment without warmup data)
- Add end_date: 2026-10-23 (3-month stopping rule) to prevent indefinite run

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 .github/workflows/smoke-copilot-sub-agents.lock.yml | 4 ++--
 .github/workflows/smoke-copilot-sub-agents.md       | 5 +++--
 2 files changed, 5 insertions(+), 4 deletions(-)

diff --git a/.github/workflows/smoke-copilot-sub-agents.lock.yml b/.github/workflows/smoke-copilot-sub-agents.lock.yml
index bdf9eb8d56b..a830e9e997e 100644
--- a/.github/workflows/smoke-copilot-sub-agents.lock.yml
+++ b/.github/workflows/smoke-copilot-sub-agents.lock.yml
@@ -1,4 +1,4 @@
-# gh-aw-metadata: {"schema_version":"v4","frontmatter_hash":"9aa2d9a657983d70b7f34ad32475efa334ec18aaefe5a4221536961e3dbfa9b8","body_hash":"66dd32a20bb7d4a839aca20b4c9729240d2574e69a5f13f3d4aaa484472974a4","strict":true,"agent_id":"copilot","agent_model":"gpt-5.3-codex","engine_versions":{"copilot":"1.0.73","copilot-sdk":"1.0.7"}}
+# gh-aw-metadata: {"schema_version":"v4","frontmatter_hash":"a36af2905338f72270a86805d6b21b53e69c06c6f7ce5a10b908bc04845f5179","body_hash":"66dd32a20bb7d4a839aca20b4c9729240d2574e69a5f13f3d4aaa484472974a4","strict":true,"agent_id":"copilot","agent_model":"gpt-5.3-codex","engine_versions":{"copilot":"1.0.73","copilot-sdk":"1.0.7"}}
 # gh-aw-manifest: {"version":1,"secrets":["COPILOT_GITHUB_TOKEN","GH_AW_GITHUB_MCP_SERVER_TOKEN","GH_AW_GITHUB_TOKEN","GITHUB_TOKEN"],"actions":[{"repo":"actions/cache/restore","sha":"55cc8345863c7cc4c66a329aec7e433d2d1c52a9","version":"v6.1.0"},{"repo":"actions/cache/save","sha":"55cc8345863c7cc4c66a329aec7e433d2d1c52a9","version":"v6.1.0"},{"repo":"actions/checkout","sha":"9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0","version":"v7.0.0"},{"repo":"actions/download-artifact","sha":"3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c","version":"v8.0.1"},{"repo":"actions/github-script","sha":"3a2844b7e9c422d3c10d287c895573f7108da1b3","version":"v9.0.0"},{"repo":"actions/setup-node","sha":"820762786026740c76f36085b0efc47a31fe5020","version":"v7.0.0"},{"repo":"actions/upload-artifact","sha":"043fb46d1a93c77aae656e7c1c64a875d1fc6a0a","version":"v7.0.1"}],"containers":[{"image":"ghcr.io/github/gh-aw-firewall/agent:0.27.38","digest":"sha256:cb928eb62d9139a013c2d278dab19af232d35a2d83dca71a3d98eb431f786243","pinned_image":"ghcr.io/github/gh-aw-firewall/agent:0.27.38@sha256:cb928eb62d9139a013c2d278dab19af232d35a2d83dca71a3d98eb431f786243"},{"image":"ghcr.io/github/gh-aw-firewall/api-proxy:0.27.38","digest":"sha256:cd6145620d96acee46e1ede25180a13aa36002467e663db0caa453a8bc8eb60c","pinned_image":"ghcr.io/github/gh-aw-firewall/api-proxy:0.27.38@sha256:cd6145620d96acee46e1ede25180a13aa36002467e663db0caa453a8bc8eb60c"},{"image":"ghcr.io/github/gh-aw-firewall/squid:0.27.38","digest":"sha256:6c19094d95aad5f9f128ad5e583f0f2b894b158aa66c3b86dd9bcc90970a2917","pinned_image":"ghcr.io/github/gh-aw-firewall/squid:0.27.38@sha256:6c19094d95aad5f9f128ad5e583f0f2b894b158aa66c3b86dd9bcc90970a2917"},{"image":"ghcr.io/github/gh-aw-mcpg:v0.4.3","digest":"sha256:3c744710ea275cd5ee65db92a1099e0d980754bd9fafda9ce67704c67004dc83","pinned_image":"ghcr.io/github/gh-aw-mcpg:v0.4.3@sha256:3c744710ea275cd5ee65db92a1099e0d980754bd9fafda9ce67704c67004dc83"},{"image":"ghcr.io/github/gh-aw-node","digest":"sha256:529d02eb970b1161aa25c593a9c3df57fdfad5a8add328cb3b6eccef66f3183b","pinned_image":"ghcr.io/github/gh-aw-node@sha256:529d02eb970b1161aa25c593a9c3df57fdfad5a8add328cb3b6eccef66f3183b"},{"image":"ghcr.io/github/github-mcp-server:v1.6.0","digest":"sha256:2b0c48b070f61e9d3969269ead600f62d00fb237b60ac849ef3d166ee7de9ad3","pinned_image":"ghcr.io/github/github-mcp-server:v1.6.0@sha256:2b0c48b070f61e9d3969269ead600f62d00fb237b60ac849ef3d166ee7de9ad3"}]}
 # This file was automatically generated by gh-aw. DO NOT EDIT. To debug this workflow, load the skill at https://github.com/github/gh-aw/blob/main/debug.md
 #
@@ -247,7 +247,7 @@ jobs:
         id: pick-experiment
         uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
         env:
-          GH_AW_EXPERIMENT_SPEC: '{"sub_agent_strategy":{"variants":["inline_strict","delegated_sequential","single_agent_control"],"description":"Measure whether inline sub-agent orchestration is the best reliability/cost tradeoff for a model-identity smoke test.","hypothesis":"H0: no change in pass_rate. H1: delegated_sequential or single_agent_control improves pass_rate by \u003e= 0.15 absolute versus inline_strict.","metric":"pass_rate","secondary_metrics":["run_duration_seconds","output_validity_rate"],"guardrail_metrics":[{"name":"empty_output_rate","direction":"min","threshold":"0.01"},{"name":"false_pass_rate","direction":"min","threshold":"0"}],"min_samples":30,"weight":[34,33,33],"issue":47551,"start_date":"2026-07-23"}}'
+          GH_AW_EXPERIMENT_SPEC: '{"sub_agent_strategy":{"variants":["inline_strict","delegated_sequential","single_agent_control"],"description":"Measure whether inline sub-agent orchestration is the best reliability/cost tradeoff for a model-identity smoke test.","hypothesis":"H0: no change in pass_rate between inline_strict and delegated_sequential. H1: delegated_sequential improves pass_rate by \u003e= 0.15 absolute versus inline_strict. Note: single_agent_control is a synthetic negative baseline (always FAIL by design) excluded from H1 comparisons.","metric":"pass_rate","secondary_metrics":["run_duration_seconds","output_validity_rate"],"guardrail_metrics":[{"name":"empty_output_rate","direction":"min","threshold":"0.01"},{"name":"false_pass_rate","direction":"min","threshold":"0.05"}],"min_samples":30,"weight":[34,33,33],"issue":47551,"start_date":"2026-07-23","end_date":"2026-10-23"}}'
           GH_AW_EXPERIMENT_STATE_FILE: /tmp/gh-aw/experiments/state.json
           GH_AW_EXPERIMENT_STATE_DIR: /tmp/gh-aw/experiments
         with:
diff --git a/.github/workflows/smoke-copilot-sub-agents.md b/.github/workflows/smoke-copilot-sub-agents.md
index 4d8543bcca0..6be47d6264d 100644
--- a/.github/workflows/smoke-copilot-sub-agents.md
+++ b/.github/workflows/smoke-copilot-sub-agents.md
@@ -20,7 +20,7 @@ experiments:
   sub_agent_strategy:
     variants: [inline_strict, delegated_sequential, single_agent_control]
     description: "Measure whether inline sub-agent orchestration is the best reliability/cost tradeoff for a model-identity smoke test."
-    hypothesis: "H0: no change in pass_rate. H1: delegated_sequential or single_agent_control improves pass_rate by >= 0.15 absolute versus inline_strict."
+    hypothesis: "H0: no change in pass_rate between inline_strict and delegated_sequential. H1: delegated_sequential improves pass_rate by >= 0.15 absolute versus inline_strict. Note: single_agent_control is a synthetic negative baseline (always FAIL by design) excluded from H1 comparisons."
     metric: pass_rate
     secondary_metrics: [run_duration_seconds, output_validity_rate]
     guardrail_metrics:
@@ -29,10 +29,11 @@ experiments:
         threshold: 0.01
       - name: false_pass_rate
         direction: min
-        threshold: 0
+        threshold: 0.05
     min_samples: 30
     weight: [34, 33, 33]
     start_date: "2026-07-23"
+    end_date: "2026-10-23"
     issue: 47551
 safe-outputs:
   create-issue: