The patch is rarely the hard part — knowing where it can land is.

landable is an Agent Skill that walks the full arc from your interests to a merged PR. It scouts repos, probes their real openness, and carries issues through to contribution — with you making only two decisions.

Install the skill

npx skills add Momoyeyu/landable -g

Requires Node.js and GitHub CLI (gh) for authenticated API access.

See how it works View on GitHub
landable workspace

The Pipeline

Six stages, two decisions

You pick what to work on and approve what gets published. Between those gates, the agent is autonomous.

1

Profile

Builds an interest card — domains, languages, layer preference — and a capability card for available hardware. Your profile shapes every downstream decision.

2

Scout

Fetches GitHub Trending within your star range, filters by interest fit, and returns a ranked candidate table with a flagged watch list just over the range.

Gate 1 — you pick repos
3

Assess

Probes each repo's real openness: who gets merged, whether issues get replies, whether the fix lives in that repo or a dependency, and whether the scope is feasible.

Gate 2 — you pick issues
4

Implement

Forks the repo, claims the issue with a comment, writes the fix and a regression test, runs the project's own test and lint gates. Nothing is pushed yet.

5

Ship

Pushes branches, opens PRs on the correct base branch respecting each repo's CONTRIBUTING.md and AGENTS.md rules, fills PR templates honestly.

6

Track

Monitors CI, diagnoses failures, follows up on maintainer review comments. Reports terminal status honestly — merged, closed-with-reason, or parked.

The Approach

What most tools skip

The interesting stage isn't implementation — it's assessment. landable checks things agents usually don't.

Acceptance probing

Samples recent merged PRs by authorAssociation. A repo where every merge is OWNER or MEMBER is a closed club — no matter how good the issues look. landable checks before you commit time.

Dependency-aware targeting

Bugs surface in one repo but root in its dependencies. The agent follows imports before committing to a target — fixing the actual cause, not the reported symptom.

Claim detection

Detects soft claims ("I'll work on this"), assignees, linked PRs, and fork references to avoid racing claimed issues. Fast-moving queues get noted, not gamed.

Rules compliance

Reads CONTRIBUTING.md, AGENTS.md, and PR templates before anything is written. Base branch, sign-off, commit signing, changelog format — the repo's own rules always win.

Results

Does it land?

Selected contributions found through landable — every link below is a merged PR.

tt-a1i/archify GitHub stars Merged

An agent skill for interactive diagrams. Repair-rounds evidence led to actionable endpoint diagnostics and clearer repair suggestions; follow-up tests closed benchmark blind spots.

PRs: #654 · #707

MakazhanAlpamys/Soup GitHub stars Merged

LLM fine-tuning on modest hardware. Scoped issues span MPS test coverage, parseable answers, evaluation scoring, numeric reward references, and false reward-hacking alarms. Validation and maintainer feedback shaped each patch.

PRs: #1375 · #1471 · #1490 · #1533 · #1630

ovg-project/kvcached GitHub stars Merged

Elastic GPU KV-cache sharing. An old listener's delayed stop could delete a replacement's live Unix socket; the patch guards unlink by socket identity and tests the replacement race.

PR: #519

Tencent-Hunyuan/UniRL GitHub stars Merged

Multimodal RL. Verified that adapter-unaware sharded-state helpers were unused before removal; separately aligned CPS rollout log-probs without changing training-side loss scaling.

PRs: #530 · #536