Agent Skills: GitHub PR Resolver

Automate GitHub pull request review resolution by processing ALL review comments and fixing ALL failing CI checks. Use when: (1) A PR has review comments that need to be addressed and resolved, (2) CI/CD checks are failing and need fixes, (3) You want to process all PR feedback and mark conversations as resolved, (4) You need to iterate on PR feedback quickly. IMPORTANT: This skill processes EVERY comment - no skipping allowed. Always fetches fresh data from GitHub. The only acceptable end state is zero unresolved threads. COMMIT CADENCE: Each thread gets its own commit (fix → commit → resolve → next). Never batch commits at the end.

workflowID: BillChirico/bills-claude-skills/github-pr-resolver

Install this agent skill to your local

pnpm dlx add-skill https://github.com/BillChirico/bills-claude-skills/tree/HEAD/github-pr-resolver

Skill Files

Browse the full folder contents for github-pr-resolver.

Download Skill

Loading file tree…

github-pr-resolver/SKILL.md

Skill Metadata

Name
github-pr-resolver
Description
"Resolve GitHub PR review comments and fix failing CI checks. Creates a team with teammates to process threads in parallel, tracks ALL threads across ALL pages, processes each with individual commits, marks each resolved immediately, and verifies all resolved + CI passing before completion."

GitHub PR Resolver

Process ALL PR review comments, make fixes, resolve threads immediately, and ensure CI passes.

Critical Rules

  1. Fetch ALL pages - GitHub returns max 100 items per request. Always paginate until hasNextPage: false
  2. Track with todos - Create a TaskCreate item for each unresolved thread before processing
  3. One commit per thread - Never batch fixes into a single commit
  4. Resolve immediately - Mark each thread resolved on GitHub RIGHT AFTER fixing it, not at the end
  5. CI must pass - Task is NOT complete until all CI checks are green. Fix failures and retry.
  6. USE TEAMMATES (MANDATORY) - You MUST create a team with TeamCreate, spawn teammates with the Task tool using team_name, and coordinate via the shared task list. Teammates work in parallel automatically. Do NOT use run_in_background agents.

API Access

Prefer GitHub MCP when available, fall back to gh CLI.

| Operation | MCP Tool | CLI Fallback | | -------------- | ---------------------------------------- | ------------------------- | | Read PR | mcp__github__pull_request_read | gh pr view | | Resolve thread | mcp__github__pull_request_review_write | gh api graphql mutation | | Check status | Included in PR read | gh pr checks |

Workflow

Step 1: Fetch ALL Threads (with Pagination)

1. Fetch PR details and review threads
2. Check hasNextPage - if true, fetch next page with cursor
3. Repeat until hasNextPage is false
4. Count total unresolved threads across ALL pages

MCP:

mcp__github__pull_request_read(owner, repo, pullNumber)
-> Check response for pagination, fetch additional pages if needed

CLI (GraphQL):

gh api graphql -f query='
query($owner: String!, $repo: String!, $prNumber: Int!, $cursor: String) {
  repository(owner: $owner, name: $repo) {
    pullRequest(number: $prNumber) {
      reviewThreads(first: 100, after: $cursor) {
        pageInfo { hasNextPage, endCursor }
        nodes {
          id, isResolved, path, line
          comments(first: 10) { nodes { body, author { login } } }
        }
      }
    }
  }
}' -f owner=OWNER -f repo=REPO -F prNumber=NUMBER

Step 2: Create Team and Task List

First, create a team:

TeamCreate:
  team_name: "pr-resolver-<prNumber>"
  description: "Resolve PR #<prNumber> review threads"

Then, for each unresolved thread (isResolved: false), create a task in the team's task list:

TaskCreate:
  subject: "[#] <path>:<line> - <summary> (@<author>)"
  description: |
    Thread ID: <id>
    Author: @<author>
    Link: https://github.com/<owner>/<repo>/pull/<prNumber>#discussion_r<commentId>
    Comment: <body>
    Repository: <owner>/<repo>
    PR Number: <prNumber>
    Branch: <branch>
    File: <path>
    Line: <line>
  activeForm: "Resolving <path>:<line> (@<author>)"

Building the comment link:

  • Extract commentId from the first comment's id field (the numeric portion after the last / or the databaseId)
  • Format: https://github.com/<owner>/<repo>/pull/<prNumber>#discussion_r<commentId>

GraphQL to get comment IDs:

gh api graphql -f query='
query($owner: String!, $repo: String!, $prNumber: Int!, $cursor: String) {
  repository(owner: $owner, name: $repo) {
    pullRequest(number: $prNumber) {
      reviewThreads(first: 100, after: $cursor) {
        pageInfo { hasNextPage, endCursor }
        nodes {
          id, isResolved, path, line
          comments(first: 1) {
            nodes {
              databaseId
              body
              author { login }
            }
          }
        }
      }
    }
  }
}' -f owner=OWNER -f repo=REPO -F prNumber=NUMBER

Verify with TaskList - count must match total unresolved from Step 1.

Step 3: Spawn Teammates to Process Threads in Parallel

CRITICAL: USE TEAMMATES FOR PARALLEL PROCESSING

You MUST spawn teammates using the Task tool with the team_name parameter. Teammates automatically work in parallel, pick up tasks from the shared task list, and send you messages when they complete or need help.

Workflow:

1. Group threads by file (threads in the same file go to one teammate)
2. Spawn one teammate per file group using Task tool with team_name
3. Each teammate independently:
   a. Claims and starts task: TaskUpdate(taskId, owner: "<teammate-name>", status: "in_progress")
   b. Reads file, makes fix
   d. git add [file] && git commit -m "[type]([scope]): [description]"
   e. Resolves thread on GitHub immediately
   f. Verifies isResolved: true
   g. Marks completed: TaskUpdate(taskId, status: "completed")
   h. Checks TaskList for more unassigned tasks
   i. Sends message to team lead when all assigned tasks are done
4. Receive automatic messages from teammates as they complete
5. Handle any failures by messaging the teammate or spawning a new one

Spawning teammates (ALL in a SINGLE message for parallel execution):

When you have 3 files to fix, spawn THREE teammates in ONE message:

YOUR SINGLE RESPONSE MUST CONTAIN:
+--------------------------------------------------------------+
|  Task #1: subagent_type="general-purpose"                    |
|           team_name="pr-resolver-<prNumber>"                 |
|           name="resolver-utils"                              |
|           description="Fix PR threads in src/utils.ts"       |
|           prompt="[teammate instructions for file 1]"        |
+--------------------------------------------------------------+
|  Task #2: subagent_type="general-purpose"                    |
|           team_name="pr-resolver-<prNumber>"                 |
|           name="resolver-api"                                |
|           description="Fix PR threads in src/api.ts"         |
|           prompt="[teammate instructions for file 2]"        |
+--------------------------------------------------------------+
|  Task #3: subagent_type="general-purpose"                    |
|           team_name="pr-resolver-<prNumber>"                 |
|           name="resolver-models"                             |
|           description="Fix PR threads in src/models.ts"      |
|           prompt="[teammate instructions for file 3]"        |
+--------------------------------------------------------------+

Teammate prompt template:

IMPORTANT: Before spawning teammates, read the team config file at ~/.claude/teams/pr-resolver-<prNumber>/config.json to get your own leader name. Inject that name into the prompt template below as [leaderName].

You are a teammate on the "pr-resolver-<prNumber>" team.
Your job is to fix PR review threads and resolve them.

Repository: [owner]/[repo]
PR Number: [prNumber]
Branch: [branch]
Team Lead: [leaderName]

Your assigned threads (all in the same file):

Thread 1:
- Task ID: [taskId]
- Thread ID: [threadId]
- File: [path]
- Line: [line]
- Comment: [body]
- Author: @[author]

[...repeat for each thread in this file...]

Instructions for EACH thread:
1. Claim task: TaskUpdate(taskId: "[taskId]", owner: "<your-name>", status: "in_progress")
2. Read the file and understand the context
3. Make the fix requested in the comment
4. Commit: git add [path] && git commit -m "[type]([scope]): [description]"
5. Resolve thread on GitHub using gh CLI:
   gh api graphql -f query='mutation { resolveReviewThread(input: {threadId: "[threadId]"}) { thread { isResolved } } }'
6. Verify resolution succeeded
7. Mark completed: TaskUpdate(taskId: "[taskId]", status: "completed")

After all threads are done:
8. Check TaskList for any remaining unassigned tasks you can pick up
9. Send a message to the team lead reporting completion:
   SendMessage(type: "message", recipient: "[leaderName]", content: "All threads resolved for [path]", summary: "Completed [path] threads")

Parallelization rules:

  • ALL teammate spawns MUST be in ONE message - This is the ONLY way to start them concurrently
  • Threads in the same file: send to a single teammate to avoid conflicts
  • Each teammate commits and resolves independently
  • Teammates go idle between turns - this is normal, they are waiting for input
  • Messages from teammates are delivered to you automatically

WRONG (sequential - do NOT do this):

Message 1: Task(team_name=...) for file A
Message 2: Task(team_name=...) for file B  <- waits for A to finish first
Message 3: Task(team_name=...) for file C  <- waits for B to finish first

CORRECT (parallel - do THIS):

Message 1: Task for file A + Task for file B + Task for file C  <- all start concurrently

Resolve thread (MCP):

mcp__github__pull_request_review_write(owner, repo, pullNumber, threadId, action: "RESOLVE")

Resolve thread (CLI):

gh api graphql -f query='
mutation($threadId: ID!) {
  resolveReviewThread(input: {threadId: $threadId}) {
    thread { id, isResolved }
  }
}' -f threadId="<THREAD_ID>"

Verify resolution succeeded:

gh api graphql -f query='
query($threadId: ID!) {
  node(id: $threadId) {
    ... on PullRequestReviewThread { isResolved }
  }
}' -f threadId="<THREAD_ID>"

Step 4: Monitor Teammates and Push Changes

Teammates send messages automatically when they finish:

1. Wait for messages from all teammates (delivered automatically)
2. Check TaskList - all todos should be "completed"
3. If a teammate reports failure, message them with guidance or spawn a replacement
4. When all tasks are complete, push all commits

Push all commits:

git push origin $(gh pr view <PR> --json headRefName -q '.headRefName')

Handle conflicts (rare with file-based grouping):

# If push fails due to conflicts from parallel commits
git pull --rebase origin <branch>
git push origin <branch>

Step 5: Wait for CI and Fix Failures

This is a loop - repeat until all checks pass:

1. Wait for CI checks to complete (not just "running")
2. Check status of ALL checks
3. If ANY check fails:
   a. Identify the failure (lint, test, build, types, etc.)
   b. Fix the issue
   c. Commit with appropriate message
   d. Push
   e. Go back to step 1
4. Only proceed when ALL checks show "success" or "skipped"

Check CI status:

# Wait for checks to complete (poll until no "pending" or "in_progress")
gh pr checks <PR> --watch

# Or check status directly
gh pr checks <PR> --json name,state,conclusion

Fix patterns for common CI failures:

# Lint failures
npm run lint -- --fix && git add . && git commit -m "fix(lint): resolve linting errors"

# Type errors
# Fix the type issues in code
git add . && git commit -m "fix(types): resolve TypeScript errors"

# Test failures
# Fix failing tests or update assertions
git add . && git commit -m "fix(tests): update failing test assertions"

# Build failures
# Fix build issues
git add . && git commit -m "fix(build): resolve build errors"

After each fix, push and wait again:

git push origin $(gh pr view <PR> --json headRefName -q '.headRefName')
# Then loop back: wait for CI, check results

Step 6: Shutdown Team and Final Verification

Only proceed here when Step 5 confirms all CI checks pass.

Shutdown all teammates:

For each teammate:
  SendMessage(type: "shutdown_request", recipient: "<teammate-name>", content: "All work complete, shutting down")
Wait for all teammates to confirm shutdown.

Clean up the team:

TeamDelete  (removes team and task directories)

Verify ALL of the following:

  1. Zero unresolved threads - Re-fetch ALL pages (paginate until hasNextPage: false):

    gh api graphql -f query='...' # Same query as Step 1
    # Count threads where isResolved: false - must be 0
    
  2. All todos completed - TaskList shows all items with status completed

  3. All CI checks passing:

    gh pr checks <PR> --json name,conclusion | jq 'all(.conclusion == "success" or .conclusion == "skipped")'
    # Must return true
    

If verification fails:

  • Unresolved threads remain -> Go back to Step 3
  • CI checks failing -> Go back to Step 5
  • Todos incomplete -> Review what was missed

Commit Convention

| Comment Pattern | Commit Type | | ------------------------------- | ----------- | | Bug fix, null check, validation | fix | | Add, implement, missing | feat | | Rename, refactor, change X to Y | refactor | | Documentation, comments | docs | | Performance | perf | | Style, formatting | style |

Scope: Extract from path - src/services/User.ts -> services

Completion Checklist

The task is NOT complete until ALL boxes can be checked:

  • [ ] Fetched ALL pages (paginated until hasNextPage: false)
  • [ ] Created team with TeamCreate
  • [ ] Created task for each unresolved thread
  • [ ] Spawned teammates for independent files in parallel (single message, multiple Task calls with team_name)
  • [ ] All teammates completed successfully (confirmed via messages and TaskList)
  • [ ] Each thread was resolved on GitHub immediately after fixing
  • [ ] All todos show completed
  • [ ] Each thread has its own commit
  • [ ] Changes pushed
  • [ ] ALL CI checks are passing (success or skipped, no failures)
  • [ ] Re-verified: zero unresolved threads across ALL pages
  • [ ] Re-verified: CI status shows all green
  • [ ] Teammates shut down with SendMessage shutdown_request
  • [ ] Team cleaned up with TeamDelete

DO NOT mark task complete if CI is still running or failing. Wait and fix.