Finn-loop skill
Three Claude Code skills turn Linear and GitHub into a software factory: spec, build and review, with a human merging.
- 324 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The agent writes code, opens PRs and changes Linear and GitHub labels on the user's behalf.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/finn-loopDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent writes code, opens PRs and changes Linear and GitHub labels on the user's behalf.
This is third-party code. Review the repository files before installing.
What it does
Finn-loop is three Claude Code skills. `/finn-spec` researches the repository, interviews the user and files a Linear issue with acceptance criteria and explicit non-goals. `/finn-build` claims the next `agent-ready` issue, implements it and opens a PR. `/finn-review` checks open PRs against the issue and required GitHub checks, then posts a verdict. Only a human applies the `agent-ready` label, and the merge also stays with a human. Build and review run in loops with `/loop /finn-build` and `/loop /finn-review`.
Who it is for. Founders and small teams that track work in Linear, host code on GitHub and want an agent to handle routine tickets under supervision.
Good fit when
- An idea needs a precise spec and the agent should ask questions before filing a Linear issue.
- A queue of small tickets waits for an agent to turn them into PRs.
- Open PRs should be checked against acceptance criteria and CI.
- The team wants a fixed rule that agents never merge code.
Not a fit when
- Work is tracked outside Linear or the code is not on GitHub.
- No required CI checks exist, and a verdict without them must not be trusted.
- You want a fully autonomous factory with no human on the merge.
Example request
Run /finn-spec for a CSV report export feature and ask questions until the behavior is unambiguous.Limitations
Needs Claude Code 2.1.71 or newer, the Linear connector and an authenticated `gh` with write access to the repository. Run one builder loop per Linear team, because two sessions under the same account cannot lock each other. `/loop` works only while the Claude Code session stays open. The skills do not create CI for the project. Extensions such as Slack and a watchdog are called ideas in the README and are absent from the repository. The last commit dates from July 2026. One build pass handles a single issue, so the queue clears gradually.
How to disable. Delete the three `finn-*` folders from `.claude/skills/` and stop any running `/loop`. Remove the Linear and GitHub labels by hand.
FAQ
Who merges the PR?
A human. Agents never merge or enable auto-merge; the `loop-approved` label is evidence for the decision only.
What does `loop-approved` mean?
The reviewer found no must-fix issue against the issue, all required GitHub checks passed and the PR has no conflicts. It informs a human and does not authorize an agent.
How do I run the builder on a schedule?
Use `/loop /finn-build`. Run review in a second session with `/loop /finn-review`.
What if the project has no CI?
There is no automated `loop-approved` verdict. Finn-loop escalates the PR for human review.
Are labels set up in advance?
Yes, but you need not create them by hand. The install prompt idempotently creates `agent-ready` and `blocked` in Linear and `loop-approved`, `loop-changes-requested` and `needs-human-review` in GitHub.
Discussion
No questions or reviews yet.
Related
SkillsA Claude Code marketplace from Intercom's Fin team: skill review, GitHub Actions hardening, permission analysis, flaky tests and pull requests
A skill for Codex and Claude Code that turns ideas and bugs into working code through a writer, tester and reviewers
Software factory installer for Claude Code and Codex: a GitHub Issues queue, skills, gates and merges done only by a human.
A plugin for Claude Code: you read one spec and parallel agents ship it to a commit with green tests.
LoopForge Local adds a role-based dev workflow to Claude Code, Codex and Cursor: analysis, design, code, review and tests.
Goal Workflow
goal-workflow
19 skills for Claude Code: PRD, spec, GitHub Issues, batch implementation with review and pull request merging, plus refactoring and diagrams
README in short
The smallest useful factory: three skills, one approval label and a human-merges rule. The README presents further layers as next steps.
README badge
Original SKILL.md
--- name: finn-spec description: Interview the user about a raw idea until confident, then file a build-ready issue in Linear. Use when asked to run Finn-loop's spec interview, draft a queue-ready issue, or plan a feature. Interactive — requires the user present; never run unattended. ---