GitHub PR Lister

Leaf skill — lists all PRs authored by the current user since a given date across one or more GitHub repos.

Inputs

• since_date: ISO date string (e.g., "2026-01-01"). Required.
• repos: List of GitHub repos (e.g., ["CXEPI/cvi-ldos-ai"]). If empty, query all repos in the org.
• org: GitHub org. Default: CXEPI.
• author: GitHub username. Default: current authenticated user (gh api user --jq .login).

Scope

In scope: Listing PR metadata (number, title, body, dates, review status, URL) for a given author and date range. Out of scope: PR content diffs, review comments, CI status. The caller handles filtering and synthesis.

Prerequisites

gh auth status

If this fails, stop and tell the user: "Run gh auth login to authenticate the GitHub CLI."

Workflow

Step 1 — Identify repos

If repos not provided:

gh repo list ORG --limit 500 --json name --jq '.[].name'

Step 2 — Fetch PRs

For each repo:

gh pr list --repo ORG/REPO_NAME --author AUTHOR --state all --search "created:>=SINCE_DATE" --limit 500 --json number,title,body,mergedAt,createdAt,reviewDecision,url

Step 3 — Compile and return

- repo: ORG/repo-name
  pr: #123
  title: "PR title"
  created: 2026-01-15
  merged: 2026-01-16
  review_decision: APPROVED
  url: https://github.com/CXEPI/repo-name/pull/123

Pagination

Single authoritative rule: Never return partial results. Fetch ALL PRs in the date range.

• Use --limit 500 per repo.
• If result count == limit, double the limit and re-fetch. Max ceiling: 2000.
• If 2000 is still hit, bisect the date range (split since_date to midpoint, fetch each half, merge).
• For org-wide queries, iterate every repo returned in Step 1.

Validation (before returning)

1. Confirm no repo was silently skipped without an explicit warning in output.
2. Confirm total PR count == sum of per-repo counts.
3. If any repo hit the 2000 ceiling even after bisection, flag it as a warning.

Error Handling

Rules

• Include both merged and unmerged PRs — caller decides what to filter.
• Do not summarize or filter — return raw metadata.