feat(ci): add autonomous CI fix loop with background monitoring

.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "github",
-  "version": "1.3.1",
-  "description": "GitHub CI/CD automation plugin for auto-detecting, analyzing, and fixing CI/CD failures on any branch",
+  "version": "1.4.0",
+  "description": "GitHub CI/CD automation plugin with autonomous fix loops, PR workflows, and code review",
   "author": {
     "name": "Ladislav Martincik",
     "url": "https://github.com/iamladi"
@@ -24,7 +24,9 @@
     "review-comments",
     "autonomous",
     "ai-review",
-    "confidence-scoring"
+    "confidence-scoring",
+    "ci-loop",
+    "auto-fix"
   ],
   "commands": {
     "fix-ci": {
@@ -46,6 +48,10 @@
   },
   "agents": [
     "./agents/ci-log-analyzer.md",
-    "./agents/ci-error-fixer.md"
+    "./agents/ci-error-fixer.md",
+    "./agents/ci-monitor.md"
+  ],
+  "skills": [
+    "./skills/ci-fix-loop/SKILL.md"
   ]
 }

CHANGELOG.md

@@ -5,6 +5,44 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [1.4.0] - 2025-12-13
+
+### Added
+
+- **Autonomous CI Fix Loop** (`/fix-ci --loop` or `/fix-ci --auto`)
+  - Runs up to 10 fix-commit-push-wait cycles automatically
+  - Fully autonomous operation with no user prompts
+  - Background CI monitoring (polls every 60 seconds)
+  - Detailed history report on completion
+  - Early abort if same errors appear twice consecutively
+
+- **New `ci-monitor` agent**
+  - Lightweight haiku-based CI status poller
+  - Token-efficient (~50 tokens per poll vs ~500 for sonnet)
+  - Runs in background to free terminal
+  - Returns: SUCCESS, FAILURE, CANCELLED, or TIMEOUT
+
+- **New `ci-fix-loop` skill**
+  - Orchestrates the full autonomous loop
+  - State tracking: attempt count, error history, run IDs
+  - Smart progress detection to avoid infinite loops
+  - Comprehensive final reporting
+
+### Safety Features
+
+- Blocks on main/master branches (must use feature branch)
+- Stashes uncommitted changes before starting
+- Maximum 30 minute wait per CI run
+- Maximum 10 retry attempts
+- Detailed commit history for easy revert
+
+### Token Efficiency
+
+- ~6,000 tokens per iteration (vs ~15,000 without optimization)
+- Haiku model for CI polling saves ~90% on wait cycles
+- No context accumulation between iterations
+- Background execution frees terminal for other work
+
 ## [1.3.1] - 2024-11-16
 
 ### Fixed

agents/ci-monitor.md

@@ -0,0 +1,94 @@
+---
+name: ci-monitor
+description: Lightweight CI status poller for background monitoring. Uses haiku model for token efficiency.
+tools: Bash
+model: haiku
+---
+
+# CI Monitor Agent
+
+Minimal CI status checker. Poll GitHub Actions and report status.
+
+## Input
+
+- `branch`: Branch name to monitor
+- `run_id`: Optional specific run ID (if known)
+
+## Behavior
+
+1. Check status:
+```bash
+gh run list --branch "${BRANCH}" --limit 1 --json databaseId,status,conclusion,createdAt
+```
+
+2. Parse and return ONE of:
+   - `QUEUED` - Run queued, not started
+   - `IN_PROGRESS` - Run executing
+   - `SUCCESS` - Completed successfully
+   - `FAILURE` - Completed with failures
+   - `CANCELLED` - Run cancelled
+   - `TIMEOUT` - Exceeded max wait (30 min)
+
+3. If `QUEUED` or `IN_PROGRESS`: wait 60 seconds, poll again
+4. If terminal state: return immediately with result
+
+## Polling Loop
+
+```bash
+MAX_WAIT=1800  # 30 minutes
+START=$(date +%s)
+POLL_INTERVAL=60
+
+while true; do
+  RESULT=$(gh run list --branch "$BRANCH" --limit 1 --json databaseId,status,conclusion 2>/dev/null)
+
+  STATUS=$(echo "$RESULT" | jq -r '.[0].status // "unknown"')
+  CONCLUSION=$(echo "$RESULT" | jq -r '.[0].conclusion // "null"')
+  RUN_ID=$(echo "$RESULT" | jq -r '.[0].databaseId // "unknown"')
+
+  case "$STATUS" in
+    completed)
+      case "$CONCLUSION" in
+        success) echo "SUCCESS|$RUN_ID"; exit 0 ;;
+        failure) echo "FAILURE|$RUN_ID"; exit 1 ;;
+        cancelled) echo "CANCELLED|$RUN_ID"; exit 2 ;;
+        *) echo "FAILURE|$RUN_ID"; exit 1 ;;
+      esac
+      ;;
+    queued|waiting|pending)
+      # Still waiting to start
+      ;;
+    in_progress)
+      # Running
+      ;;
+    *)
+      # Unknown status, keep polling
+      ;;
+  esac
+
+  ELAPSED=$(($(date +%s) - START))
+  if [ $ELAPSED -gt $MAX_WAIT ]; then
+    echo "TIMEOUT|$RUN_ID"
+    exit 3
+  fi
+
+  sleep $POLL_INTERVAL
+done
+```
+
+## Output Format
+
+Single line: `STATUS|RUN_ID`
+
+Examples:
+- `SUCCESS|12345678`
+- `FAILURE|12345678`
+- `CANCELLED|12345678`
+- `TIMEOUT|12345678`
+
+## Important
+
+- Be extremely concise
+- No explanations - just return status
+- Max poll time: 30 minutes
+- Poll interval: 60 seconds

commands/fix-ci.md

@@ -9,13 +9,56 @@ Auto-detect, analyze, and fix CI/CD failures on any branch using GitHub CLI and
 ## Usage
 
 ```bash
-/fix-ci              # Current branch
-/fix-ci 123          # PR number
-/fix-ci https://...  # PR URL
+/fix-ci              # Current branch (single fix)
+/fix-ci 123          # PR number (single fix)
+/fix-ci https://...  # PR URL (single fix)
+/fix-ci --loop       # Autonomous loop mode (up to 10 retries)
+/fix-ci --auto       # Alias for --loop
+/fix-ci 123 --loop   # Loop mode for specific PR
 ```
 
 User provided: `$ARGUMENTS`
 
+## Autonomous Loop Mode
+
+When `--loop` or `--auto` flag is present, this command runs in autonomous mode using the `ci-fix-loop` skill.
+
+**Detection:**
+```bash
+ARGS="$ARGUMENTS"
+if [[ "$ARGS" == *"--loop"* ]] || [[ "$ARGS" == *"--auto"* ]]; then
+  # Extract PR number if provided (e.g., "123 --loop" → "123")
+  PR_NUM=$(echo "$ARGS" | grep -oE '^[0-9]+' || echo "")
+
+  # Invoke ci-fix-loop skill
+  # The skill will handle the full autonomous loop:
+  # 1. Analyze CI errors
+  # 2. Apply fixes
+  # 3. Commit and push
+  # 4. Monitor CI in background (polling every 60s)
+  # 5. If CI fails, repeat (up to 10 times)
+  # 6. Report final status
+
+  # IMPORTANT: Do not proceed with single-fix workflow below
+fi
+```
+
+**Behavior:**
+- Runs up to 10 fix-commit-push-wait cycles
+- Fully autonomous (no user prompts)
+- Background CI monitoring between iterations
+- Reports detailed history when complete
+- Aborts if same errors appear twice consecutively
+
+**Safety:**
+- Will not run on main/master branch
+- Stashes uncommitted changes before starting
+- Maximum 30 minute wait per CI run
+
+---
+
+When NOT in loop mode, proceed with single-fix workflow below.
+
 ## Workflow
 
 ### Step 1: Detect Context