peterruiter/claude-build-framework

Write an agent brief

Template: ${CLAUDE_PLUGIN_ROOT}/templates/agent-brief.md. Background: ${CLAUDE_PLUGIN_ROOT}/docs/waves-and-agents.md.

Steps

1. Gather facts (do not guess):
   • repo path, branch, stack summary (CLAUDE.md), gate commands (CLAUDE.md or the profile);
   • languages: read build/contracts/locales.json: count, source, the longest language (usually German or Finnish) and a plural-heavy one (one whose Intl.PluralRules needs few/many, if configured) for screenshots;
   • data residency (CLAUDE.md);
   • strings folder, glossary, changelog (highest id: say "next free at the moment you write it");
   • reserved migration slot, scratch DB, ports, scratch folder (from the wave plan, or pick free ones and record them);
   • the files to study first and the files the agent owns.
2. Write the goal in user terms, then concrete scope bullets and what is out of scope.
3. Fill every placeholder of the template. Remove sections that truly do not apply (e.g. screenshots for a backend-only item) and say why in one line.
4. Keep the standing rules verbatim from the template: all configured languages with plural forms, WCAG 2.2 AA, light and dark, privacy by design, data residency, one What's new entry, STATE.md section, small targeted edits, no commit/push/deploy/cloud, no secrets, no AI attribution in anything the agent writes.
5. Checks with numbers and the report format at the end.
6. If the item touches security-sensitive surfaces (sign-in, tokens, links, webhooks, exports, tenant boundaries), add the relevant lines from ${CLAUDE_PLUGIN_ROOT}/docs/security-review.md as explicit acceptance criteria with tests.
7. Output the brief as one block, ready to pass as the agent's prompt.