feat: split triage and pr-fixer into separate GHA workflows#1177
Merged
Gkrumbach07 merged 2 commits intomainfrom Apr 2, 2026
Merged
feat: split triage and pr-fixer into separate GHA workflows#1177Gkrumbach07 merged 2 commits intomainfrom
Gkrumbach07 merged 2 commits intomainfrom
Conversation
- triage.yml: daily cron (8am UTC weekdays), discovers untriaged Jira/GH issues, creates Implement sessions (max 5/cycle) - pr-fixer.yml: 30 min cron, manages all ai-managed PRs with session reuse, change detection, and circuit breaker - fix-single: triggered only by @ambient-fix comment - fix-batch: scheduled + manual dispatch - Both use ambient-action@v0.0.3 with inactivity timeout (60s) - Unified label: ai-managed (was agent-managed) - Removed old matrix-based batch pattern - Skip Jira gracefully if MCP tool unavailable Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Contributor
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (2)
🚧 Files skipped from review as they are similar to previous changes (1)
📝 WalkthroughWalkthroughPR fixer workflow reworked to run fixes from issue comments and a single orchestrator batch job using session-frontmatter prompts and Claude Opus; batching now uses an orchestrator prompt with retry/circuit-breaker logic. Added a scheduled Issue Triage workflow that queries Jira/GitHub and spawns child investigation/fix sessions. Changes
Sequence Diagram(s)sequenceDiagram
participant IssueComment as Issue Comment Event
participant GitHubAPI as GitHub API
participant PRBody as PR Body Parser
participant AmbientAction as ambient-action v0.0.3
participant Claude as Claude Opus Model
participant SessionMgmt as Session Management
IssueComment->>GitHubAPI: event payload (issue_comment)
GitHubAPI-->>PRBody: fetch PR/issue body
PRBody->>PRBody: extract acp:session_id and frontmatter
PRBody-->>AmbientAction: invoke fix-single with session context
AmbientAction->>Claude: run session-driven fix prompt
Claude->>SessionMgmt: read/create session state
Claude-->>AmbientAction: return fix operations & updated frontmatter
AmbientAction->>GitHubAPI: apply PR body updates/comments
GitHubAPI-->>AmbientAction: confirm update
sequenceDiagram
participant Schedule as Scheduler (cron/dispatch)
participant AmbientAction as ambient-action v0.0.3
participant Jira as Jira (MCP)
participant GitHub as GitHub Issues
participant Claude as Claude Opus Model
participant ChildSession as Child Session Creator
Schedule->>AmbientAction: trigger triage job
AmbientAction->>Jira: query untriaged items (max 5)
alt Jira available
Jira-->>AmbientAction: return items
else Jira unavailable
AmbientAction->>GitHub: query untriaged issues
GitHub-->>AmbientAction: return items
end
AmbientAction->>Claude: run triage orchestrator prompt
Claude->>GitHub: check for ai-managed PRs / existing sessions
Claude->>ChildSession: create investigation/fix child sessions as needed
ChildSession-->>Claude: return session name & UID
Claude->>GitHub: add `ai-triaged` label and comment with session link
GitHub-->>AmbientAction: confirm updates
AmbientAction->>Schedule: write session summary to step summary
🚥 Pre-merge checks | ✅ 6✅ Passed checks (6 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
✨ Simplify code
Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 7
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In @.github/workflows/pr-fixer.yml:
- Around line 50-61: The existing session lookup step (id: existing) writes
session_id and source to GITHUB_OUTPUT but those outputs are never consumed;
update later steps to read steps.existing.outputs.session_id and
steps.existing.outputs.source where session continuity is needed, and stop
unconditionally resetting retry_count to 0 — instead only initialize retry_count
when no session_id was found (e.g., if: steps.existing.outputs.session_id ==
''), or pass the previous retry_count through the workflow when a session
exists; in short, wire the existing step's outputs into the logic that decides
session continuation and remove the hard-reset of retry_count so an `@ambient-fix`
resumes the prior session/retry state.
- Around line 8-13: The scheduled workflow can overlap with long-running runs;
update the workflow to prevent concurrent runs by adding a top-level concurrency
block (e.g., concurrency: group: "pr-fixer-${{ github.ref }}"
cancel-in-progress: false) so new triggers (from schedule or workflow_dispatch)
will queue until the current run finishes, or alternatively change the schedule
cron to an interval >= the max run time (e.g., hourly) to avoid overlap; apply
this change near the existing schedule and workflow_dispatch entries so all
triggers are serialized.
- Around line 141-173: The workflow currently increments retry_count
unconditionally after sending a message ("After sending: increment
retry_count..."), which treats normal churn as failure; change this so
retry_count is incremented only when the agent fails to make progress (no new
external commits, no new/updated review comments, CI still failing, merge
conflicts unresolved) or when the agent explicitly reports a failure state, and
reset or leave retry_count unchanged when the agent's action produced a
meaningful change; update the logic around the session management step that
references retry_count/last_action/ai-needs-human (the "After sending" block and
any code that updates frontmatter like <!-- acp:session_id=... last_action=...
retry_count=... -->) to base increments on a success flag or a comparison of
repo state before/after the agent run rather than incrementing unconditionally.
- Around line 68-71: The workflow currently uses the mutable tag "uses:
ambient-code/ambient-action@v0.0.3" in two places (the "Fix PR" step and the
other step later) which weakens supply-chain security; update both "uses:
ambient-code/ambient-action@v0.0.3" occurrences to pin the action to the
corresponding full commit SHAs (replace `@v0.0.3` with @<commit-sha> for each
occurrence), ensuring you fetch the exact commit SHAs from the
ambient-code/ambient-action repository and update both instances consistently.
- Around line 131-133: The gh CLI call that lists PRs uses the default of 30
results; update the Run command that invokes `gh pr list --repo ${{
github.repository }} --state open --label ai-managed --search "draft:false"` to
include an explicit limit (e.g. add `--limit 1000`) so all matching PRs are
returned; modify the `gh pr list` invocation to `gh pr list ... --search
"draft:false" --limit 1000` (or another appropriate numeric limit) to ensure no
PRs are silently omitted.
In @.github/workflows/triage.yml:
- Around line 3-10: Add a top-level concurrency guard to the triage workflow to
serialize scheduled and manual runs: insert a concurrency block (e.g.,
concurrency: group: 'triage-${{ github.workflow }}' cancel-in-progress: false)
so that runs triggered by schedule and workflow_dispatch cannot overlap and will
queue instead of racing; update the .github/workflows/triage.yml file around the
existing on: schedule and workflow_dispatch sections to include this concurrency
configuration.
- Around line 20-23: The workflow step named "Run triage orchestrator" (id:
session) currently pins the action as ambient-code/ambient-action@v0.0.3;
replace the tag with the specific full commit SHA for
ambient-code/ambient-action (e.g.,
ambient-code/ambient-action@<full-commit-sha>) so the action is immutably
pinned; update the uses value in that step accordingly and commit the change.
🪄 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: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
Run ID: ae0efb6c-68d1-4b14-bfb5-4c0fbac18bc4
📒 Files selected for processing (2)
.github/workflows/pr-fixer.yml.github/workflows/triage.yml
.github/workflows/pr-fixer.yml
Outdated
Comment on lines
+50
to
+61
| - name: Check for existing session | ||
| if: steps.fork_check.outputs.skip != 'true' | ||
| id: existing | ||
| env: | ||
| GH_TOKEN: ${{ secrets.GITHUB_TOKEN }} | ||
| run: | | ||
| # Read PR body and extract session_id from frontmatter | ||
| BODY=$(gh pr view ${{ steps.pr.outputs.number }} --repo "${{ github.repository }}" --json body --jq '.body') | ||
| SESSION_ID=$(echo "$BODY" | grep -oP 'acp:session_id=\K[^ ]+' | head -1 || echo "") | ||
| SOURCE=$(echo "$BODY" | grep -oP 'source=\K[^ ]+' | head -1 || echo "") | ||
| echo "session_id=$SESSION_ID" >> $GITHUB_OUTPUT | ||
| echo "source=$SOURCE" >> $GITHUB_OUTPUT |
Contributor
There was a problem hiding this comment.
The existing-session lookup currently has no effect.
Line 60 stores session_id, but nothing later reads steps.existing.outputs.session_id. Line 89 also hard-resets retry_count=0, so an @ambient-fix on an already-managed PR loses its prior session/retry state instead of continuing it.
Also applies to: 81-90
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In @.github/workflows/pr-fixer.yml around lines 50 - 61, The existing session
lookup step (id: existing) writes session_id and source to GITHUB_OUTPUT but
those outputs are never consumed; update later steps to read
steps.existing.outputs.session_id and steps.existing.outputs.source where
session continuity is needed, and stop unconditionally resetting retry_count to
0 — instead only initialize retry_count when no session_id was found (e.g., if:
steps.existing.outputs.session_id == ''), or pass the previous retry_count
through the workflow when a session exists; in short, wire the existing step's
outputs into the logic that decides session continuation and remove the
hard-reset of retry_count so an `@ambient-fix` resumes the prior session/retry
state.
Comment on lines
68
to
+71
| - name: Fix PR | ||
| if: steps.fork_check.outputs.skip != 'true' | ||
| id: session | ||
| uses: ambient-code/ambient-action@v0.0.2 | ||
| uses: ambient-code/ambient-action@v0.0.3 |
Contributor
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
cd /dev/null 2>&1; pwd > /dev/null; cat -n ".github/workflows/pr-fixer.yml" | head -130Repository: ambient-code/platform
Length of output: 6302
Pin both ambient-code/ambient-action references to full commit SHAs instead of @v0.0.3.
Semantic version tags are mutable and weaken supply-chain security for a workflow that modifies PR state and accesses secrets. Replace with pinned commit SHAs at lines 71 and 123.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In @.github/workflows/pr-fixer.yml around lines 68 - 71, The workflow currently
uses the mutable tag "uses: ambient-code/ambient-action@v0.0.3" in two places
(the "Fix PR" step and the other step later) which weakens supply-chain
security; update both "uses: ambient-code/ambient-action@v0.0.3" occurrences to
pin the action to the corresponding full commit SHAs (replace `@v0.0.3` with
@<commit-sha> for each occurrence), ensuring you fetch the exact commit SHAs
from the ambient-code/ambient-action repository and update both instances
consistently.
.github/workflows/pr-fixer.yml
Outdated
Comment on lines
+141
to
+173
| ## 2. Circuit breaker | ||
| If retry_count >= 3: comment "AI was unable to resolve after 3 attempts. Needs human attention.", | ||
| add ai-needs-human label, remove ai-managed label. Skip. | ||
|
|
||
| ## 3. Check for changes since last_action | ||
| Ignore commits authored by the bot. Only look for: | ||
| - New commits by someone other than the bot | ||
| - New or updated review comments | ||
| - New CI failures | ||
| - Merge conflicts from base branch changes | ||
|
|
||
| If nothing changed → skip entirely. | ||
|
|
||
| ## 4. Something changed — act | ||
| - CI failing → send message to existing session with CI logs | ||
| - New review comments → send message with the comments | ||
| - Merge conflicts → send message to rebase | ||
| - New external commits → send message to review and ensure CI passes | ||
|
|
||
| ## 5. Session management | ||
| Before sending a message, check session status: | ||
| - Running → send the message | ||
| - Stopped → restart (reuse), then send | ||
| - Not found → create new session with this prompt: | ||
| "You are maintaining an open pull request. | ||
| PR: <URL> Source issue: <KEY> (if known) | ||
| 1. Check out the PR branch. | ||
| 2. Resolve merge conflicts, fix CI failures, address review comments. | ||
| 3. Push fixes. Do not merge/close/force-push. | ||
| 4. Write frontmatter: <!-- acp:session_id=$AGENTIC_SESSION_NAME source=<KEY> last_action=<NOW> retry_count=0 --> | ||
| 5. If broken beyond repair, comment and stop." | ||
|
|
||
| After sending: increment retry_count and update last_action in frontmatter. |
Contributor
There was a problem hiding this comment.
The circuit breaker currently counts normal churn as failure.
Line 173 increments retry_count after every message, regardless of outcome. Three ordinary rounds of new comments/CI churn will trip ai-needs-human even if the agent is making progress.
Suggested prompt tweak
- After sending: increment retry_count and update last_action in frontmatter.
+ After sending:
+ - If the PR is still broken after your attempted fix, increment retry_count.
+ - If the PR is healthy again, reset retry_count to 0.
+ - Update last_action in frontmatter.📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| ## 2. Circuit breaker | |
| If retry_count >= 3: comment "AI was unable to resolve after 3 attempts. Needs human attention.", | |
| add ai-needs-human label, remove ai-managed label. Skip. | |
| ## 3. Check for changes since last_action | |
| Ignore commits authored by the bot. Only look for: | |
| - New commits by someone other than the bot | |
| - New or updated review comments | |
| - New CI failures | |
| - Merge conflicts from base branch changes | |
| If nothing changed → skip entirely. | |
| ## 4. Something changed — act | |
| - CI failing → send message to existing session with CI logs | |
| - New review comments → send message with the comments | |
| - Merge conflicts → send message to rebase | |
| - New external commits → send message to review and ensure CI passes | |
| ## 5. Session management | |
| Before sending a message, check session status: | |
| - Running → send the message | |
| - Stopped → restart (reuse), then send | |
| - Not found → create new session with this prompt: | |
| "You are maintaining an open pull request. | |
| PR: <URL> Source issue: <KEY> (if known) | |
| 1. Check out the PR branch. | |
| 2. Resolve merge conflicts, fix CI failures, address review comments. | |
| 3. Push fixes. Do not merge/close/force-push. | |
| 4. Write frontmatter: <!-- acp:session_id=$AGENTIC_SESSION_NAME source=<KEY> last_action=<NOW> retry_count=0 --> | |
| 5. If broken beyond repair, comment and stop." | |
| After sending: increment retry_count and update last_action in frontmatter. | |
| ## 2. Circuit breaker | |
| If retry_count >= 3: comment "AI was unable to resolve after 3 attempts. Needs human attention.", | |
| add ai-needs-human label, remove ai-managed label. Skip. | |
| ## 3. Check for changes since last_action | |
| Ignore commits authored by the bot. Only look for: | |
| - New commits by someone other than the bot | |
| - New or updated review comments | |
| - New CI failures | |
| - Merge conflicts from base branch changes | |
| If nothing changed → skip entirely. | |
| ## 4. Something changed — act | |
| - CI failing → send message to existing session with CI logs | |
| - New review comments → send message with the comments | |
| - Merge conflicts → send message to rebase | |
| - New external commits → send message to review and ensure CI passes | |
| ## 5. Session management | |
| Before sending a message, check session status: | |
| - Running → send the message | |
| - Stopped → restart (reuse), then send | |
| - Not found → create new session with this prompt: | |
| "You are maintaining an open pull request. | |
| PR: <URL> Source issue: <KEY> (if known) | |
| 1. Check out the PR branch. | |
| 2. Resolve merge conflicts, fix CI failures, address review comments. | |
| 3. Push fixes. Do not merge/close/force-push. | |
| 4. Write frontmatter: <!-- acp:session_id=$AGENTIC_SESSION_NAME source=<KEY> last_action=<NOW> retry_count=0 --> | |
| 5. If broken beyond repair, comment and stop." | |
| After sending: | |
| - If the PR is still broken after your attempted fix, increment retry_count. | |
| - If the PR is healthy again, reset retry_count to 0. | |
| - Update last_action in frontmatter. |
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In @.github/workflows/pr-fixer.yml around lines 141 - 173, The workflow
currently increments retry_count unconditionally after sending a message ("After
sending: increment retry_count..."), which treats normal churn as failure;
change this so retry_count is incremented only when the agent fails to make
progress (no new external commits, no new/updated review comments, CI still
failing, merge conflicts unresolved) or when the agent explicitly reports a
failure state, and reset or leave retry_count unchanged when the agent's action
produced a meaningful change; update the logic around the session management
step that references retry_count/last_action/ai-needs-human (the "After sending"
block and any code that updates frontmatter like <!-- acp:session_id=...
last_action=... retry_count=... -->) to base increments on a success flag or a
comparison of repo state before/after the agent run rather than incrementing
unconditionally.
Comment on lines
+20
to
+23
| - name: Run triage orchestrator | ||
| id: session | ||
| uses: ambient-code/ambient-action@v0.0.3 | ||
| with: |
Contributor
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
cd /
find . -name "triage.yml" -type f 2>/dev/null | head -20Repository: ambient-code/platform
Length of output: 113
🏁 Script executed:
cat -n .github/workflows/triage.ymlRepository: ambient-code/platform
Length of output: 4347
🏁 Script executed:
# Check if there are any exposed secrets in the workflow
rg -n "(GITHUB_TOKEN|secrets\.|env\.)" .github/workflows/triage.yml -A 2Repository: ambient-code/platform
Length of output: 351
Pin ambient-action to a full commit SHA.
Version tags like @v0.0.3 are mutable and can be retagged, allowing behavior changes without triggering another review. Replace with the commit SHA. Secrets and permissions are correctly scoped.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In @.github/workflows/triage.yml around lines 20 - 23, The workflow step named
"Run triage orchestrator" (id: session) currently pins the action as
ambient-code/ambient-action@v0.0.3; replace the tag with the specific full
commit SHA for ambient-code/ambient-action (e.g.,
ambient-code/ambient-action@<full-commit-sha>) so the action is immutably
pinned; update the uses value in that step accordingly and commit the change.
- Add concurrency guards to prevent overlapping runs - Remove unused existing-session lookup step from fix-single - Add --limit 200 to gh pr list in batch orchestrator - Fix circuit breaker: only increment retry_count on failure, reset to 0 when PR is healthy Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
pr-fixer.ymlinto two focused workflows with distinct cadencestriage.yml— daily issue discovery and session creationpr-fixer.yml— 30 min PR management with session reuse and circuit breakerChanges
triage.yml(new)ai-triagedto issues, child sessions addai-managedto PRspr-fixer.yml(rewritten)@ambient-fixPR comment (was also label + manual)ai-managedPRslast_action(ignores bot commits)ai-needs-humanlabelfix-batch/fix-eachpatternai-managed(wasagent-managed)Both workflows
ambient-action@v0.0.3with inactivity timeout (timeout: '60'= 60s idle auto-stop)claude-opus-4-6Test plan
triage.yml— verify it discovers issues and creates sessions@ambient-fixon a PR — verifyfix-singletriggerspr-fixer.yml(no PR number) — verify batch orchestrator runsai-managedlabel triggers batch pickup on next cycle🤖 Generated with Claude Code
Summary by CodeRabbit