Skip to content

Agent Skills

Fresh

Automated Claude Code skills for fixing PR review comments and iterating to a perfect Greptile score.

Agent skills automate the full auto-fix workflow in Claude Code. Instead of manually prompting your assistant to fetch comments and apply fixes, invoke a skill and let it handle the entire loop.

Both skills are open source and available at github.com/greptileai/skills.

Note

Prerequisites:

  • Claude Code installed
  • git and gh (GitHub CLI) installed and authenticated
  • Greptile installed on your repository
  • Greptile MCP server configured in Claude Code

Install

Clone the skills into your Claude Code skills directory. See the Claude Code skills docs for more on how skill discovery works.

bash
git clone https://github.com/greptileai/skills.git ~/.claude/skills/greptile

Or add as a submodule in your project:

bash
git submodule add https://github.com/greptileai/skills.git .skills/greptile

check-pr

Checks a pull request for unresolved review comments, failing status checks, and incomplete descriptions, then fixes and resolves them.

Usage

text
/check-pr

Or with a specific PR number:

text
/check-pr 42

If no PR number is given, the skill auto-detects the PR for your current branch.

What It Does

Waits for pending checks

Polls until all CI and review bot checks (Greptile, linters, etc.) reach a terminal state. This ensures all comments are available before analysis.

Analyzes the PR

Evaluates four areas:

* **Status checks**, Are CI checks passing or failing?
* **PR description**, Is it complete? Any TODOs or placeholders?
* **Review comments**, Inline comments from Greptile, linters, and human reviewers
* **General comments**, Discussion threads and bot notifications

Categorizes issues

Each issue is classified as:

| Category              | Meaning                                          |
| --------------------- | ------------------------------------------------ |
| **Actionable**        | Code changes, test improvements, or fixes needed |
| **Informational**     | Verification notes, questions, or FYIs           |
| **Already addressed** | Resolved by subsequent commits                   |

Fixes and resolves

If there are actionable items, the skill makes the fixes, commits, pushes, and resolves the corresponding review threads.

greploop

Iteratively improves a PR until Greptile gives it a 5/5 confidence score with zero unresolved comments. Triggers a Greptile review, fixes all actionable comments, pushes, re-triggers, and repeats.

Usage

text
/greploop

Or with a specific PR number:

text
/greploop 42

What It Does

Triggers a Greptile review

Pushes the latest changes and waits for the Greptile review check to complete.

Parses review results

Reads the confidence score (e.g. `3/5`) and fetches all unresolved inline comments from Greptile.

Fixes actionable comments

For each unresolved comment, reads the file in context, determines if a code change is needed, and applies the fix. Informational comments and false positives are noted and resolved.

Commits, pushes, and loops

Commits the fixes, pushes, and goes back to step 1. The loop runs up to **5 iterations** to avoid runaway cycles.

Exits when clean

Stops when Greptile returns **5/5 confidence** with **zero unresolved comments**, or after the max iterations.

Output

On success:

Greploop complete.
  Iterations:    2
  Confidence:    5/5
  Resolved:      7 comments
  Remaining:     0

If the loop hits the iteration limit:

Greploop stopped after 5 iterations.
  Confidence:    4/5
  Resolved:      12 comments
  Remaining:     2

Remaining issues:
  - src/auth.ts:45, "Consider rate limiting this endpoint"
  - src/db.ts:112, "Missing index on user_id column"

Next Steps

Auto-Fix Workflow

Manual auto-fix patterns using MCP tools

Claude Code Setup

Set up the Greptile MCP server for Claude Code

Source

Mirrored from the official Greptile documentation. Self-contained reference copy.