| name | taste |
| description | Auto-bootstrap project taste.md + taste.vision via /deepresearch. NO 10-question wizard. Reads repo signals + researches SOTA at current time → synthesizes taste docs with zero questions (one fallback question only if repo has no signal). |
/taste — auto-bootstrap project kernel
Replaces v1's /tastebootstrap (10 kernel questions). v2 auto-derives taste from repo evidence + current-time SOTA research.
Process
- Read repo signals: README, package manifest, top-level structure, framework detection.
- /deepresearch the project's SOTA at current time:
- " best practices "
- " SOTA architecture "
- WebFetch official docs of detected framework.
- Synthesize (Opus) →
taste.md (operating principles, code style, architecture invariants, deps posture) + taste.vision (north star, ICP, success criteria) with zero questions.
- ONE clarifying question only if the repo has no signal (empty dir, no README, no package manifest). Exact question: "What are you building, in one sentence?"
- Show the generated docs; user accepts or edits and re-runs.
Why auto-bootstrap
- 10 questions is friction. The model already has more signal from the repo + web than the user would type.
- Anchored on current-time SOTA, not training-cutoff opinions.
- Re-runnable cheaply — drift detection via /deepretaste is the same engine.
CLI
cmax taste init # auto-bootstrap; writes taste.md + taste.vision
cmax taste init --regenerate # ignore existing; re-derive from scratch
What goes in taste.md vs taste.vision
- taste.md — declarative rules. "We use X, not Y, because Z." Code style, architecture, deps.
- taste.vision — narrative north star. One paragraph each: what we're building, who for, what "great" looks like.
How the harness uses taste
- SessionStart hook reads
taste.md and injects it into every workflow's system prompt.
- /introspect cross-references plans against taste.md and flags deviations.
- /audit and /review flag taste violations in existing code and proposed diffs.
