Finn-spec: Turn Raw Ideas into Build-Ready Linear Issues

Where this fits

You tell your agent "add usage limits to the API key dashboard," and an hour later you have a Linear issue that says "do things." The missing step is a spec. finn-spec fixes that by researching the codebase and interviewing you until the raw idea becomes a build-ready Linear issue. You stay the product brain; the agent becomes the codebase brain. It never guesses product decisions.

Why agents benefit

• Researches relevant code before asking questions, so it only asks what the codebase cannot answer.
• Runs the interview in rounds of 1-4 questions, with concrete options and a recommended option listed first.
• Produces an issue with scope boundaries, acceptance criteria, and edge cases, so a build agent needs no extra context.
• Refuses to guess product decisions; it flags them as questions for the user, preventing silent scope drift.
• Runs interactively only, so an unattended agent can never file half-baked issues.

Practical scenarios

From bug report to queue-ready issue

A vague report like "Order export fails for CSV with empty address" is turned into a precise issue. The agent researches the export module, then asks only the behavior fork: should the export skip the row, fill a placeholder, or fail loudly? It files an issue with exact failure handling and tests to update.

New feature across UI and API

"Let users archive projects" touches UI, API, permissions, and data semantics. Finn-spec checks existing patterns, then asks only the product forks: who can unarchive, what happens to active tasks, and which roles see archived projects. It files a Linear issue that a build agent can execute without revisiting product questions.

Product idea needing scope boundaries

A raw idea like "support teams" gets broken into a minimal vertical slice. The issue includes explicit out-of-scope items such as notifications and audit logs, so estimation and building stay manageable.

Add it to your agent workflow

Install with:

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

Then prompt:

Run finn-spec on this idea: users should be able to invite teammates

The agent will research, ask a few rounds of questions, draft the Linear issue, and confirm before filing. Keep a human in the loop for the whole process.

Compare before adopting

Compare finn-spec against three things. Quality signals: run it on a real idea and see whether the resulting issue reduces follow-up questions from your build agent. Maintenance freshness: check commit activity and issue responses in the Finn-loop repo. Alternatives: generic plan mode or your own spec prompt may fit better if your team needs Notion, Jira, or a custom question template. Workflow fit: this skill is Linear-specific and interactive, so make sure that matches how your team launches work.

Why it is worth tracking

Finn-spec is worth tracking because it encodes a workflow most teams only approximate in chat. The 305 GitHub stars and active refinements in Finn-loop suggest the interview format is already helping other developers. Track it when you want your agent to turn ideas into issues without hallucinating requirements. For more skill comparisons, see the finn-spec profile at OpenAgentSkill: https://www.openagentskill.com/skills/finna-finn-spec