MediaMetz/gemini-cli
1
0
Fork 0
mirror of https://github.com/google-gemini/gemini-cli.git synced 2026-05-23 02:02:31 -07:00
Code Issues Packages Projects Releases Wiki Activity

fix(bot): implement prompt-level loop prevention and enforce test timeouts

Browse Source
This commit is contained in:
Christian Gunderman
2026-05-13 21:15:05 -07:00
parent 928ce879ea
commit a2841350ba
2 changed files with 17 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/gemini-cli-bot/.gemini/skills/issue-fixer/SKILL.md
+1 -1
View File
@@ -16,7 +16,7 @@ maintain existing PRs to reduce the repository backlog.
2. **Discovery**: Find open issues labeled `effort/small`. Prioritize those with clear reproduction steps. You are STRICTLY FORBIDDEN from using local commands (e.g., `npm run lint`, `npm run typecheck`, or grepping for TODOs) to find new work.
3. **Autonomous Implementation**: You are responsible for the entire fix: research, code changes, and test verification.
4. **Surgical Precision**: Changes must be minimal and strictly focused on the identified issue. Avoid "drive-by" refactoring.
5. **Local Verification**: You MUST run `timeout 10m npm run preflight` locally and iterate on any failures before finalizing your PR. Local commands are for VERIFICATION ONLY, never for discovery.
5. **Local Verification (MANDATORY TIMEOUTS)**: You MUST run `timeout 10m npm run preflight` (or `timeout 5m npm test ...`) locally and iterate on any failures before finalizing your PR. Wrapping test commands in `timeout` is **MANDATORY** to prevent hanging the CI environment if your changes introduce an infinite loop.
6. **Expert Mentions**: Identify the domain expert for the affected files and CC them in the PR description.
7. **Focused Contributions**: Limit your active PRs to ~10 at a time. Try to complete existing PRs before opening new ones. If a maintainer closes a PR, that may be an indication that they are rejecting the fix.
tools/gemini-cli-bot/brain/scheduled.md
+16
View File
@@ -94,3 +94,19 @@ For any detected bugs, bottlenecks, or opportunities:
- **Strict Read-Only Reasoning**: You cannot push code or post comments via API.
Your only way to effect change is by writing to specific files and explicitly
staging file changes using the `git add` command.
## Loop Prevention & Success Criteria
To ensure efficiency and prevent infinite reasoning loops:
1. **Monitor Your Progress**: If you have attempted the same sequence of
actions (e.g., Edit -> Test -> Fail) for the same problem more than **3
times**, you MUST stop and re-evaluate your fundamental hypothesis.
2. **Failure Threshold**: If you cannot find a verified solution after **2
distinct hypotheses** (max 6 total edit/test cycles), you MUST abort the
task.
3. **Reporting Failure**: If you abort, summarize the roadblocks you
encountered in `lessons-learned.md`. It is better to deliver NO changes than
to burn excessive tokens on a loop.
4. **Verification is Key**: A task is only "complete" when all relevant tests
pass. Never stage a change that you know still fails tests.