A Tiny CI Guard That Lets a Free AI Model Read Your Build Failures

Description and use

The guard processed over a hundred classification requests in a week without hitting limits, correctly identifying missing dependencies, permission issues, and syntax errors, cutting triage time from ~10 to ~2 minutes.

Problem

CI pipelines generate unreadable logs; developers waste time locating the real error and suggesting fixes.

Outcome

The guard processed over a hundred classification requests in a week without hitting limits, correctly identifying missing dependencies, permission issues, and syntax errors, cutting triage time from ~10 to ~2 minutes.

Workflow

• 1. Configure MonkeyCode endpoint and API key. 2. Run Node.js script that watches CI output file. 3. Extract last 200 lines of failing step plus project name and branch. 4. Send extracted log to MonkeyCode OpenAI‑compatible endpoint with a system prompt for CI log analysis. 5. Receive JSON with errorType, confidence, suggestedFix. 6. Apply decision table: if confidence ≥0.80 show fix and auto‑open file; if 0.50–0.79 show error type and ask for confirmation; if <0.50 show raw log excerpt. 7. Retry up to three times on malformed JSON, stripping markdown fences. 8. Output single‑line summary (e.g., "[type] confidence=0.83 fix=Add null check before calling .map").