Brief

Clarify only missing product decisions: user, problem, scope, constraints and observable success. Do not repeat information already supplied or require a numeric business metric for a small usability change. Group related questions in plain prose; no start/proceed confirmations. -a/-s allow explicit assumptions, not invented certainty or external permissions.

Write docs/brief/<slug>/brief.md and tasks.md from references/brief-template.md, never from a brief found in another project. Keep its frontmatter and structural headings Acceptance criteria, Success metrics, Out-of-scope, Boundary, Relevant Files and Tasks. Derive outcome tasks from acceptance criteria; do not inflate them into a fixed number of phases. One feature, one folder.

Boundary names what this feature owns and what it must not touch, in two short lists of paths. It is what makes two briefs safe to run at the same time, so write it whenever the answer is knowable and leave it out rather than guess. It bounds the work, not the reader: it is not a file inventory.

Mark ready only when the brief is complete; never implement automatically. At finalize (steps/step-04-finalize.md), record the Git base and branch when the directory is a repository, and offer one issue per task through the issue skill when gh is available; both are optional and a brief without them still ships. Render with scripts/render.py only for --html or an explicit request. No mandatory browser, commit, test run, or handoff menu. Report paths and the next action if requested. Existing files are updated, not duplicated.

Arsenal handoff

When called by Arsenal for an explicitly requested broader task, return the brief, its readiness and unresolved decisions to Arsenal. Do not launch ship yourself. Arsenal may continue an authorized implementation once the brief is ready; a request for a brief alone never authorizes implementation.

Verify external facts

Before stating an external fact (a version, an API, a price, a date, a regulation, the behaviour of a tool or library, a person or company), search first when the runtime offers web search: read the websearch skill and run one quick search, then cite the source. One search per fact; --deep only on request. Without web search, mark the claim unverified. Facts visible in the repository or the request need no search.

Execution policy

Work solo. Ask before any subagent or reviewer, even in auto mode. Explain the independent scope and expected benefit first. No hidden advisor, nested delegation, model retuning, repeated successful checks, or progress spam. Use existing context before asking questions. Stop when the requested result is delivered. User stops and scope changes override pending steps.