report: a receipt for the work

Turn a completed cycle into a concise, skimmable record someone can read without re-reading the diff. It's the same content as the in-chat verdict, made durable and shareable.

When it runs

• On request: "give me a report", "write up what you did", "summarize this cycle".
• Automatically at the end of build / qa only if the project preference says so (below). Default is off: otherwise the in-chat verdict is the whole deliverable.

Preference (per project): off by default

Read .mastermind/prefs.md in the project root for a cycle-report: line: one of:

• off (default, and the assumption if the file/key is missing): report only on explicit request.
• ask: at the end of a report-worthy cycle, offer once: "want a report? markdown / html / no".
• markdown: always write a Markdown report.
• html: always write a self-contained HTML report.

The init skill sets this once, at the first task. The user can change it anytime ("reports off", "always give me html reports"): update the line in .mastermind/prefs.md to match.

What goes in it: signal, not a log

1. Title + verdict: Done · Done, not fully checked · Not done · Wrong thing, one line why, and the not checked: line.
2. What changed: files touched (path + one-line what), grouped by area.
3. Key decisions: the non-obvious calls and the one-line reason (not what the model already knows).
4. How it was verified: what was actually run/observed end-to-end, and what it produced.
5. Follow-ups / risks: anything left, edge cases, TODOs.

Keep it tight: a bloated report gets ignored (same rule as the brain: signal density beats volume).

Format

• Markdown (default): write .mastermind/reports/<slug>-<YYYY-MM-DD>.md (use today's date), or a path the user names. Cheap, diffable, opens anywhere.
• HTML (on request / cycle-report: html): one self-contained file (inline CSS, no external assets or scripts), lightly MasterMind-styled, so it opens in any browser. Costs more tokens.

Tool-agnostic: always a plain file on disk, so it survives whatever tool wrote it.

Cost & restraint

An HTML report adds meaningful output tokens (~1.5–4k) vs Markdown (~0.4–1k); the in-chat verdict is ~free. Match effort to stakes: for a one-line change, skip the report entirely: offer nothing. Reserve it for cycles someone would actually want to read or share.