await-greptile

star 2

Autonomous PR review agent that monitors Greptile, fixes issues, and merges when approved. Runs in background after PR creation. Loops until 5/5 confidence score then auto-merges. Use after creating a PR for fully hands-off review workflow.

roach88 By roach88 schedule Updated 2/23/2026

name: await-greptile description: Autonomous PR review agent that monitors Greptile, fixes issues, and merges when approved. Runs in background after PR creation. Loops until 5/5 confidence score then auto-merges. Use after creating a PR for fully hands-off review workflow.

Await Greptile - Autonomous PR Review Agent

A background agent that monitors Greptile code reviews, automatically fixes any issues found, and merges when the review passes with high confidence. Runs autonomously until complete.

Trigger

Use this skill when:

  • Just created a PR and want fully autonomous review handling
  • Want hands-off PR workflow: create → fix → merge
  • User invokes /await-greptile or /await-greptile <PR#>

Arguments

  • <PR#> — Optional PR number. If omitted, detects from current branch.
  • --no-merge — Don't auto-merge, just fix issues and report
  • --max-iterations=N — Maximum fix/review cycles (default: 5)

Workflow

Step 1: Gather Context

  1. Get repo details from git remote:

    git remote get-url origin | sed 's/.*github.com[:/]\(.*\)\.git/\1/'
    
  2. If PR number not provided, detect from current branch:

    git branch --show-current
    

    Then use mcp__plugin_greptile_greptile__list_pull_requests with sourceBranch filter.

  3. Get full PR context via mcp__plugin_greptile_greptile__get_merge_request:

    • PR title and description (THE INTENT)
    • Changed files list
    • Current review status
    • Existing comments

Step 2: Load Agent Persona

Read the agent definition from:

~/.claude/skills/await-greptile/AGENT.md

This contains:

  • Issue categorization strategies
  • Fix approaches for each category
  • Safety rails
  • Output formats

Step 3: Launch Background Agent

Use the Task tool to spawn the autonomous agent:

Task(
  subagent_type: "general-purpose",
  run_in_background: true,
  description: "PR Review Agent for #{pr_number}",
  prompt: """
  <agent-context>
  {contents of AGENT.md}
  </agent-context>

  <pr-context>
  PR: #{pr_number}
  Title: {pr_title}
  Repo: {owner/repo}
  Branch: {branch_name}

  ## Original PR Description (THE INTENT)
  {pr_description}

  ## Files Changed
  {list of changed files with paths}

  ## Configuration
  Max iterations: {max_iterations}
  Auto-merge: {true/false}
  </pr-context>

  Begin the review loop now. Start by checking the current review status.
  """
)

Step 4: Notify User

Immediately after launching:

🤖 Autonomous PR Review Agent started

**PR:** #{pr_number} - {title}
**Task ID:** {task_id}

The agent will:
1. ⏳ Wait for Greptile to review
2. 🔧 Fix any issues found (up to {max_iterations} cycles)
3. ✅ Merge when 5/5 confidence achieved

**Check status:** TaskOutput(task_id: "{task_id}", block: false)

Continue working - the PR will be handled automatically.

Agent Behavior Reference

The spawned agent follows the full protocol in AGENT.md, including:

Issue Categories

  1. Type/Schema Issues — Fix type mismatches, branded type access
  2. Missing Validation — Add Zod schemas or runtime guards
  3. Error Handling — Add try/catch, meaningful errors
  4. Performance Issues — Fix N+1, add caching, optimize
  5. Test Quality — Replace placeholders with real assertions
  6. Code Style — Extract constants, improve naming
  7. Security Issues — Sanitize inputs, never commit secrets
  8. Architecture — Fix boundary violations, circular deps

Safety Rails

  • Max 5 iterations (prevents infinite loops)
  • Tests run before every push
  • Specific file adds only (never git add .)
  • No force pushes
  • Stops on security issues for human review
  • Always references original PR intent

Decision Points

  • 5/5 confidence + no issues → Merge and stop
  • Issues found → Fix, push, wait for re-review
  • Max iterations → Stop without merging
  • Test failure → Stop and report
  • Ambiguous/security issue → Stop for human review

Example Output

🤖 Autonomous PR Review Agent started

**PR:** #42 - Fix Probability API mismatch
**Task ID:** task_abc123

The agent will:
1. ⏳ Wait for Greptile to review
2. 🔧 Fix any issues found (up to 5 cycles)
3. ✅ Merge when 5/5 confidence achieved

Continue working - the PR will be handled automatically.

---

[Background agent output - check with TaskOutput]

## Iteration 1/5

**Status:** COMPLETED
**Confidence:** 4/5

### Issues Found: 2
| # | Category | File | Issue | Status |
|---|----------|------|-------|--------|
| 1 | Type | test.ts:42 | .value on branded number | ✅ Fixed |
| 2 | Test | spec.ts:10 | Missing assertion | ✅ Fixed |

### Tests: PASSED ✅
### Commit: abc123

**Next:** Waiting for re-review...

---

## Iteration 2/5

**Status:** COMPLETED
**Confidence:** 5/5
**Issues Found:** 0

### 🎉 Merging PR #42...

---

## PR Review Complete

**PR:** #42 - Fix Probability API mismatch
**Final Status:** MERGED ✅
**Iterations:** 2
**Fixes Applied:** 2
**Time:** 4m 32s

Manual Invocation

# Full autonomous loop
/await-greptile

# Specific PR
/await-greptile 42

# Don't auto-merge
/await-greptile 42 --no-merge

# Limit iterations
/await-greptile --max-iterations=3
Install via CLI
npx skills add https://github.com/roach88/claude-config --skill await-greptile
Repository Details
star Stars 2
call_split Forks 0
navigation Branch main
article Path SKILL.md
More from Creator