fix(autoresearch): add GitHub Issue escalation as step 7 in stuck detection#183
fix(autoresearch): add GitHub Issue escalation as step 7 in stuck detection#183maystudios merged 1 commit intomainfrom
Conversation
…ection Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
Pull request overview
This PR updates the autoresearch “When Stuck” protocol to add an explicit escalation path via creating a diagnostic GitHub Issue, intended to prevent infinite recovery loops and require human guidance after repeated failures.
Changes:
- Adds a new Step 7 (“Escalate”) in the “When Stuck” section.
- Documents a
gh issue createcommand template with labels/title/body placeholders for diagnostics.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| 4. Try combining 2-3 previously successful changes. | ||
| 5. Try the opposite of what has not been working. | ||
| 6. Try a radical architectural change. | ||
| 7. **Escalate** — If still stuck after all above: create a diagnostic GitHub Issue (`gh issue create --label "type:bug" --label "maxsim:auto" --title "Stuck: [scope] after [N] consecutive failures" --body "[full context: iterations attempted, approaches tried, current metric value, TSV log summary, suspected blockers]"`) and escalate to the user. Do not continue the loop — wait for human guidance. |
There was a problem hiding this comment.
This new escalation step instructs the agent to stop and wait for human guidance, but earlier in this same document “Unbounded Mode (default)” says “Never stop. Never ask if the agent should continue.” Please clarify that escalation is an explicit exception to Unbounded Mode (e.g., update the Unbounded Mode text or reference the exception here) to avoid contradictory instructions.
| 4. Try combining 2-3 previously successful changes. | ||
| 5. Try the opposite of what has not been working. | ||
| 6. Try a radical architectural change. | ||
| 7. **Escalate** — If still stuck after all above: create a diagnostic GitHub Issue (`gh issue create --label "type:bug" --label "maxsim:auto" --title "Stuck: [scope] after [N] consecutive failures" --body "[full context: iterations attempted, approaches tried, current metric value, TSV log summary, suspected blockers]"`) and escalate to the user. Do not continue the loop — wait for human guidance. |
There was a problem hiding this comment.
The title template includes “after [N] consecutive failures”, but this section is specifically triggered by “5 consecutive discards or crashes”. Consider making the title/wording consistent with the trigger (e.g., use 5, or 5+), so the created Issue clearly reflects the actual stuck threshold being applied.
| 7. **Escalate** — If still stuck after all above: create a diagnostic GitHub Issue (`gh issue create --label "type:bug" --label "maxsim:auto" --title "Stuck: [scope] after [N] consecutive failures" --body "[full context: iterations attempted, approaches tried, current metric value, TSV log summary, suspected blockers]"`) and escalate to the user. Do not continue the loop — wait for human guidance. | |
| 7. **Escalate** — If still stuck after all above: create a diagnostic GitHub Issue (`gh issue create --label "type:bug" --label "maxsim:auto" --title "Stuck: [scope] after 5 consecutive discards or crashes" --body "[full context: iterations attempted, approaches tried, current metric value, TSV log summary, suspected blockers]"`) and escalate to the user. Do not continue the loop — wait for human guidance. |
|
🎉 This PR is included in version 5.13.2 🎉 The release is available on: Your semantic-release bot 📦🚀 |
Summary
Test plan
🤖 Generated with Claude Code