decisions-logger

Scaffolds a project's context stack — the tiered docs folders that let an agent be briefed on a project — and registers the routing table in CLAUDE.md.

Reviews a built interface against its stated design intent — naming how the thing reads before proposing a fix, ranking findings by severity, and treating removal as a first-class verdict. Refuses to critique without a written intent.

Documents every substantive commit into per-commit changelog files plus a rolling CHANGELOG index.

Creates and maintains docs/MODEL-STRATEGY.md — which Claude model to use for which work, tailored via interview.

Interviews you into a written visual direction — the surface and its brand posture, the references and the anti-reference, the constraints as breakable rules, and the vocabulary — and asks what you want rather than proposing a look for you.

Proposes three or four sourced angles worth posting from your material backlog and the record — each with its tension, its reader and its weakest point. Declines when nothing is worth a post yet.

Generates a new skill in this library's conventions — PRD, SKILL.md skeleton, plugin.json, marketplace entry, README row and website entry — from a short interview.

Generates three genuinely distinct design directions from a written visual direction — each committed to a named structural axis, each with its thesis, bet and risk — ending in a comparison matrix you judge by eye. Never records the verdict.

Splits the released version from the build you're testing, checks every file carrying the version agrees, and keeps a generated ledger of every build and release — across npm, Claude Code plugins, Docker, PyPI, binaries, sites, extensions, mobile apps and plain exports.

Interviews you into a short, structured design brief — problem, who feels it, jobs to be done, constraints, success criteria and non-goals — and marks what wasn't stated rather than inventing it.