Finn-build: Safe Agent Loop for Shipping Linear Issues

Where this fits

Your agent is running in a /loop, and each iteration should either fix review feedback or ship one issue. Without guardrails, it might grab a blocked task, dirty your working tree, or push to the wrong branch. finn-build turns that into a single safe pass: claim the next approved Linear issue, implement it, and open a PR.

Why agents benefit

• Preflight before mutation: checks the repo, detects the real default branch, and requires a clean git status --porcelain before changing anything.
• Review feedback first: picks up PRs labeled loop-changes-requested and skips anything carrying needs-human-review.
• One pass = one unit of work: each loop iteration is measurable, idempotent, and safe to repeat.
• Linear as source of truth: agents work from an approved queue instead of inventing tasks.
• Built for /loop: the skill is scoped so a single invocation can be reviewed and stopped if needed.

Practical scenarios

Fix review feedback without unrelated changes

The agent lists open PRs with loop-changes-requested, inspects the latest review comments, implements fixes on the existing branch, and updates the PR. If the working tree is dirty, it reports the paths and ends the pass.

Build the next approved Linear issue

With no review feedback pending, the agent claims the next safe agent-ready issue, creates a branch from the detected default branch, implements the change, and opens a PR referencing the issue.

Handle an empty queue gracefully

When there are no ready Linear issues and no PRs needing changes, the agent reports that there is nothing to do and exits cleanly instead of fabricating work.

Add it to your agent workflow

Install the skill with:

npx skills add finna/Finn-loop --skill finn-build

Then use it inside a loop prompt:

Run /loop with finn-build one pass: fix any loop-changes-requested PR, otherwise build one approved Linear issue and open a PR.

The skill's preflight will stop early if origin is unreachable, the default branch cannot be detected, or the working tree is not clean.

Compare before adopting

Before wiring it in, compare:

• Quality signals: Does it verify branch state, PR labels, and review queues instead of blindly generating code?
• Maintenance freshness: Check Finn-loop's recent commits and open issues; Linear and GitHub workflows need active updates.
• Alternatives: Generic autonomous coding agents may be broader, but finn-build is opinionated about the Linear-to-PR loop.
• Workflow fit: You need Linear, GitHub, and labels like loop-changes-requested or you will have to adapt the flow.

Why it is worth tracking

finn-build is worth tracking because it encodes a safe, single-turn agent loop that respects both an external review queue and a clean tree. With 305 stars and a focused README, it signals real community interest in constrained agent workflows. Evaluate it when your team has an approved Linear backlog and wants agents to ship small units of work without manual supervision. For more context, see the skill page.