awais786/.claude/skills/pick-issue

You are helping a Python/Django expert find the best open source issues to contribute to.

User Profile

See profile.md for expertise, preferred categories, and repos/labels to skip.

Ecosystem Detection

Check $ARGUMENTS for ecosystem hints:

• If it contains ai, AI, llm, LLM → fetch AI repos directly (see Step 1b)
• If it contains python → focus on py_core, py_web, py_tools categories
• Default → Django preferred categories from profile.md

Steps

Step 0 — Skip Already-Worked Issues

Check the ## Contribution History section in MEMORY.md (loaded automatically into context). Any issue URL listed there must be discarded immediately — do not recommend, score, or verify it.

Step 1a — Hub Issues (Django/Python ecosystem)

Cache path: /Users/awais.qureshi/Documents/devstack/opensource-issues/.claude/skills/pick-issue/cache/issues.json Cache TTL: 24 hours

Check if cache exists and is fresh:

CACHE=/Users/awais.qureshi/Documents/devstack/opensource-issues/.claude/skills/pick-issue/cache/issues.json
if [ -f "$CACHE" ]; then
  AGE=$(( $(date +%s) - $(stat -f %m "$CACHE" 2>/dev/null || stat -c %Y "$CACHE") ))
  [ $AGE -lt 86400 ] && echo "fresh" || echo "stale"
else
  echo "missing"
fi

• If fresh and $ARGUMENTS does NOT contain fresh → read from cache file, skip download
• If stale, missing, or $ARGUMENTS contains fresh → fetch from URL, save to cache:

mkdir -p /Users/awais.qureshi/Documents/devstack/opensource-issues/.claude/skills/pick-issue/cache

Fetch: https://awais786.github.io/opensource-issues/data/issues.json Then write the result to the cache path above.

Step 1b — Fetch AI Issues (only when ecosystem = "ai")

Always fetch live (GitHub API data changes frequently):

• https://api.github.com/repos/run-llama/llama_index/issues?state=open&per_page=20&sort=created&direction=desc
• https://api.github.com/repos/BerriAI/litellm/issues?state=open&per_page=20&sort=created&direction=desc
• https://api.github.com/repos/chroma-core/chroma/issues?state=open&per_page=15&sort=created&direction=desc
• https://api.github.com/repos/langchain-ai/langchain/issues?state=open&labels=good+first+issue&per_page=15
• https://api.github.com/repos/modelcontextprotocol/python-sdk/issues?state=open&per_page=15&sort=created&direction=desc

Filter out pull requests (items with pull_request field).

Step 2 — Filter

Apply from profile.md:

• Skip repos in skip list
• Skip issues with excluded labels
• Also skip labels: question, invalid, wontfix, stale, duplicate

If $ARGUMENTS specifies a category or filter, apply it on top.

Step 3 — Score

• +3 if is_good_first_issue: true
• +2 if is_help_wanted: true
• +3 if comments == 0 (no competition)
• +2 if comments < 5
• +1 if updated within last 14 days
• +1 if is_bug: true
• +1 if priority: "low" or "medium"
• -1 if comments > 15
• -2 if created_at more than 1 year ago

Step 4 — Verify Top Candidates (always, no exceptions)

Take the top 15 scored issues and verify each via GitHub API in parallel:

https://api.github.com/repos/<owner>/<repo>/issues/<number>

Discard an issue if ANY of these are true:

• state is not "open"
• assignees array is non-empty (someone claimed it)
• pull_request field exists in the response (it's a PR not an issue)

Check for linked PRs explicitly:

https://api.github.com/repos/<owner>/<repo>/issues/<number>/timeline

If any event has event: "cross-referenced" and the source is an open pull request → discard.

Only proceed with issues confirmed open + unassigned + no linked PRs.

Step 5 — Present Top 5 (verified clean only)

| # | Repo | Issue | Type | Comments | Score | Link | |---|------|-------|------|----------|-------|------|

All issues in this table are guaranteed: open, unassigned, no in-progress PRs.

Step 6 — Top Recommendation

Single best pick with: why it fits your skills, concrete approach, estimated scope (small/medium/large), and direct GitHub link.

To implement: say "Make a PR for <issue-url>"