oss-contributor
Skill0 starsChanged 6 months agoMITView on GitHub
Use this skill whenever the user wants to find, claim, track, or contribute to open-source GitHub issues — especially in AI/ML repositories like lancedb, BerriAI/litellm, run-llama/llama_index, weaviate, or deepset-ai/haystack. Trigger on phrases like "find OSS issues", "good first issue", "unclaimed GitHub issue", "I want to contribute to open source", "find me a PR to work on", or the Chinese equivalents "找开源 issue", "认领 issue", "贡献开源", "找 PR", "我想给 X 提 PR". Also use this skill when the user asks about tracking claim status, deadline reminders, PR monitoring after submitting a contribution, mentions they want to become an AI Engineer through OSS contributions, asks "is this issue still available?", or says it's their first time contributing to open source — even when they don't explicitly mention this skill by name.
- Reads credentials
- Installs packages
1 other project uses a version of this.
Adopted by
Most recently taken up by
- Edzard9/stashtheir own version, first seen Nov 16, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- Edzard9/stash★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 1 is distinctive, and 6 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 12
Use this skill whenever the user:
- Very specific: line 21
If the user's project doesn't have .claude/oss-config.json yet, treat that as a strong signal they need Step 1 (first-time setup) before …
- Very specific: line 38
Check whether .claude/oss-config.json exists in the user's current project.
- Very specific: line 68
When the user picks an issue, help them post a claim comment on GitHub.
- Very specific: line 70
After they confirm the comment is posted, append to active_claims in .claude/oss-config.json:
- Very specific: line 153
All scripts use only the Python standard library — no pip install required.
- Very specific: line 161
All scripts must be run from the user's project root (the directory containing .claude/), not from the skill directory.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
OSS Contributor
A skill for systematically finding, claiming, and tracking open-source GitHub contributions — with a focus on AI/ML repositories.
When to use this skill
Use this skill whenever the user:
- Asks to find unclaimed GitHub issues, "good first issues", or contribution opportunities
- Wants to contribute to open source, especially AI/ML projects
- Wants to track issues they've already claimed (status, deadline, PR feedback)
- Asks "is this issue still available?" or whether an issue was claimed
- Says they want to grow into an AI Engineer role via OSS contributions
- Asks how to write a good claim comment, or wants their first OSS PR merged
If the user's project doesn't have .claude/oss-config.json yet, treat that as a strong signal they need Step 1 (first-time setup) before anything else.
Core principle: never trust GitHub search results alone
GitHub's web UI and search index make issues appear unclaimed when they're not. Cases that naive search misses and this skill catches:
- Hidden linked PRs — Cross-referenced PRs (especially closed ones) often don't show on the issue page, but the Timeline API reveals them
- Recent claim comments — Search indexing lags by minutes; a comment posted 10 minutes ago saying "I'll work on this" may not appear in search results yet
- JS-rendered comments — HTML scraping can't see them; the REST Comments API can
- Closed-but-not-merged PRs — These signal "someone tried, gave up"; usually worth skipping
That's why scripts/oss-issue-finder.py runs a 5-point verification against the live GitHub REST API for every candidate before recommending it. The algorithm is documented in references/verification-algorithm.md — read that file if the user asks why a specific issue was filtered out or wants to tune the criteria.
Workflow
Step 1 — First-time setup (once per project)
Check whether .claude/oss-config.json exists in the user's current project. If not, run:
bash scripts/oss-init.shIt interactively asks for:
- GitHub username
- Target repos (Enter for defaults:
lancedb/lancedb,BerriAI/litellm,weaviate/weaviate-python-client,run-llama/llama_index,deepset-ai/haystack) - Difficulty levels (
1easy /2medium /3hard, or combos like1,2) - Days to complete each issue (default: 7)
- Daily tracking time in HH:MM 24h format (default: 08:00)
The script writes config to .claude/oss-config.json in whatever directory it was run from. All other scripts read this same relative path, so always invoke them from the user's project root.
Step 2 — Find unclaimed issues
python3 scripts/oss-issue-finder.py <difficulty><difficulty> is 1, 2, 3, or a comma-separated combo like 1,2. Returns up to 5 verified-unclaimed issues, ranked by recency + simplicity.
GitHub rate limit gotcha. Unauthenticated API allows 60 requests/hour. A full search across 5 repos with verification typically consumes 30–60 requests. If the user hits the limit:
- Wait one hour
- Reduce
target_reposin.claude/oss-config.json - Set the
GITHUB_TOKENenvironment variable (raises limit to 5000/hour) — only if the user already has a token; don't generate or ask for one unsolicited
Step 3 — Help the user claim an issue
When the user picks an issue, help them post a claim comment on GitHub. Read references/claim-comment-templates.md first — it has two styles (humble vs. standard) and guidance for choosing. Match what the user asks for ("谦逊一点" → humble; default → standard).
After they confirm the comment is posted, append to active_claims in .claude/oss-config.json:
{
"issue_id": "lancedb-3153",
"repo": "lancedb/lancedb",
"number": 3153,
"title": "<issue title>",
"claim_date": "<ISO timestamp now>",
"deadline": "<claim_date + days_to_complete>",
"status": "claimed",
"pr_url": null
}The status field progresses through this lifecycle:
claimed → in_progress → at_risk (≤3 days left) → pr_open → merged (or pr_closed / overdue / abandoned).
The tracker script updates status automatically; you only set it to claimed at creation time.
Step 4 — Track claims
python3 scripts/oss-claim-tracker.pyThis is meant to run daily. Wire it up either via Claude's /schedule skill or a cron job — see references/scheduling-integration.md for quota math and three pre-tuned scheduling profiles.
For each active claim it verifies:
- The user's claim comment still exists on the GitHub issue (matched by username + keyword like "i'll work", "claiming", "i'd like to")
- Whether a linked PR exists, and if so, its state (open / merged / closed)
- Days remaining until the deadline
Escalation pattern (this is why the deadline field matters):
- Days remaining > 3:
in_progress, no nag - Days remaining ≤ 3:
at_risknotification - Past deadline:
overduenotification - Comment vanished:
abandoned(user either deleted it or never had matching keywords)
Step 5 — When the PR merges
Move the entry from active_claims to completed in .claude/oss-config.json. Add:
{
"issue_id": "...",
"repo": "...",
"pr_url": "https://github.com/.../pull/123",
"merged_date": "<ISO timestamp>"
}The tracker will detect the merge state automatically, but moving it out of active_claims keeps the daily run focused on still-open work.
Common situations and what to do
"I want to contribute to open source but don't know where to start"
→ Check for .claude/oss-config.json. If missing, run oss-init.sh. Then oss-issue-finder.py 1 for the easiest issues. Walk through the top result with them.
"What's the status of my claims?"
→ Run oss-claim-tracker.py. It prints a report and updates statuses. Highlight anything at_risk or overdue.
"How do I automate this daily?"
→ Read references/scheduling-integration.md. Recommend the balanced profile (~40 tasks/month) unless they say otherwise. Use the /schedule skill to wire it up.
The finder returns zero results → Likely: rate-limit hit (wait 1h), too-strict difficulty, or the easy ones genuinely got claimed in the last few hours by other people running the same kind of script. Try a different difficulty or remove one repo from the config.
"My claim comment isn't being detected by the tracker"
→ The tracker matches on both username AND a keyword. Keywords it looks for: i'll work, i'll take, assign me, claiming, i can fix, i'd like to work, can i work, i would like, i'm interested, i am interested. If the user wrote something else, suggest editing the comment to include one of those phrases — or pull from references/claim-comment-templates.md.
"I want to add a new target repo"
→ Edit target_repos in .claude/oss-config.json directly. No script needed; the next oss-issue-finder.py run picks it up.
Reference files — when to read which
The references below are loaded only when needed, so you don't burn context on details that aren't relevant to the current request.
references/verification-algorithm.md— Read when the user asks why an issue was filtered out, when they want to tune the criteria, or when they're debugging "this looked unclaimed on the website but the script skipped it"references/claim-comment-templates.md— Read any time the user is about to post a claim comment. Has humble version, standard version, and follow-up message examplesreferences/scheduling-integration.md— Read when the user asks about automating the daily/weekly runs, hits a quota concern, or wants to compare aggressive vs. balanced vs. deep profiles
Script reference
All scripts use only the Python standard library — no pip install required.
| Script | Purpose | Args | Reads | Writes |
scripts/oss-init.sh | One-time interactive setup | none (prompts) | — | .claude/oss-config.json |
scripts/oss-issue-finder.py | Find unclaimed issues | <difficulty> (1/2/3/1,2) | .claude/oss-config.json | stdout |
scripts/oss-claim-tracker.py | Verify claims, update statuses | none | .claude/oss-config.json | same config + .claude/logs/oss-workflow.log |
All scripts must be run from the user's project root (the directory containing .claude/), not from the skill directory.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

