Load PR Comments

Load Unresolved PR Review Comments as Parallel Tasks

Load ONLY open/UNRESOLVED PR review threads and rewrite them into grouped markdown task files under .specs/comments/*.md, each safe for a separate parallel agent to implement with no overlap.

Critical Guidelines

• You MUST load ONLY threads where the resolved state is false. Skip resolved threads.
• You MUST rewrite each comment as an actionable TASK requirement, not a summary. Preserve substance (code suggestions, exact instructions) verbatim.
• You MUST NOT post, reply to, or modify anything on GitHub. This skill is read-only against the API.
• You MUST group comments so each file is independently implementable with NO duplication across files.

Step 0: Verify what tools are available

1. Check if GitHub CLI is installed and authenticated:

   gh auth status

2. Check if the GitHub MCP server is available:

   mcp__MCP_DOCKER__pull_request_read

   or simular command without MCP_DOCKER prefix, if installed directly.

• if both are available, use any that have enough accesses to the repository.
• if github mcp server avaiable but have different structure, adjust in order to fit the expected structure.
• if none is available, try to load directly through curl in case if it is public repository. If it is private, ask user to install GitHub CLI or GitHub MCP server.
• if MCP server not installed, but github cli is installed, but not authenticated, ask user to run gh auth login to authenticate.

Step 1: Resolve the Target PR

• An explicit PR argument ALWAYS takes precedence over current-branch resolution. If a PR number or URL was passed, use it (a URL like https://github.com/{owner}/{repo}/pull/{n} → number {n}) and do NOT consult the current branch.
• Otherwise default to the PR of the CURRENT branch:

gh pr view --json number,url,headRefName   # current branch's PR

• Resolve repo owner/name: gh repo view --json owner,name.
• If no PR exists for the branch, gh pr view errors with "no pull requests found" — STOP and report that no PR is associated with the branch (ask for a PR number/URL).

Step 2: Retrieve UNRESOLVED Comments

Resolved/unresolved is a GraphQL reviewThreads { isResolved } concept — the REST /pulls/{n}/comments endpoint does NOT expose it. Use one of the two approaches below; if the primary is unavailable, fall back to the other.

Primary: gh CLI (GraphQL)

Filter isResolved == false directly in the jq:

gh api graphql -f query='
query($owner:String!,$repo:String!,$pr:Int!){
  repository(owner:$owner,name:$repo){
    pullRequest(number:$pr){
      reviewThreads(first:100){
        nodes {
          isResolved
          isOutdated
          path
          line
          startLine
          originalLine
          comments(first:50){
            nodes { author{login} body diffHunk url }
          }
        }
      }
    }
  }
}' -F owner=OWNER -F repo=REPO -F pr=PR_NUMBER \
  --jq '[.data.repository.pullRequest.reviewThreads.nodes[]
         | select(.isResolved==false)
         | {path, isOutdated,
            line: (.line // .startLine // .originalLine),
            comments: [.comments.nodes[]
             | {author: .author.login, url, body, diffHunk}]}]'

This returns each unresolved thread with its file path, an isOutdated flag, a usable line, and ordered comments (author, body, permalink url, diffHunk).

IMPORTANT: For OUTDATED threads line is null (the diff moved). The jq above already falls back line // startLine // originalLine, so line is never null when any anchor exists. When ALL three are null, omit the :<line> segment entirely in the template — never render path:null.

Fallback: GitHub MCP

If the GitHub MCP server (MCP_DOCKER) is available, use the mcp__MCP_DOCKER__pull_request_read tool with method: "get_review_comments":

mcp__MCP_DOCKER__pull_request_read
  method: "get_review_comments"
  owner: "OWNER"
  repo: "REPO"
  pullNumber: PR_NUMBER
  perPage: 100

It returns review_threads[], each with is_resolved, is_outdated, is_collapsed (all snake_case — verified against this tool's response) and comments[] (body, path, author, html_url). Keep only threads where is_resolved is false. Note the MCP comment objects expose path but NOT a line number, so use the html_url as the location anchor. Paginate with after: <endCursor> while pageInfo.hasNextPage is true.