neural-claude-code

Render a Claude Code plan (PlanViewer), a task result, or a recap of a branch/commit/PR diff (ResultViewer) as a single-file HTML artifact with anti-fatigue UX. Use when the user asks for an HTML artifact, a plan/result/recap view, or to "show this as HTML". Not for plain markdown/text or one-liners.

Build a planned change: take an approved /spec plan, capture a baseline, execute the subtasks, review with /vet and /exercise in a clean context, measure the delta, and stop for ship. If no approved plan exists, runs /spec first. Don't use for trivial changes (<25 LOC); edit directly.

Behavioral verification gate: run the test suite, then drive the running software as a real user (CLI/HTTP by default; browser or computer-use if those MCP tools are available) to confirm a feature works end to end. Reports PASS/FAIL per scenario with evidence. Use after a build, before ship, or as /craft's behavioral gate alongside /vet. Don't use for static code review (that is /vet).

Stage and commit code changes with Conventional Commits format. Use when: code changes are complete and ready to commit, after review approval, user says 'save', 'commit', 'ship it'. Don't use when: tests are failing or work is still in progress.

Scan a codebase for accumulated slop, technical debt, dead code, and refactor opportunities. Use when user says 'scan for slop', 'find tech debt', 'what needs cleanup', or before a refactor sprint.

Plan a non-trivial change BEFORE any code: research the context, decompose into testable subtasks, lock signatures, note CWE invariants, write executable acceptance, and surface contradictions. Produces an approvable plan artifact and STOPS for review. Does not write code. Pairs with /craft, which executes the plan. Don't use for trivial changes (<25 LOC) or pure conversation.

Clean-context review gate. An independent reviewer (a fresh Agent, with no author context) challenges the diff for correctness, verifies it meets the approved spec's acceptance criteria, and checks project consistency. Verdict SHIP/HOLD/BLOCK. Use standalone before a PR/merge, or as craft's review phase. Don't use for trivial changes (<15 LOC) or docs-only edits.

Scan project and generate a customized CLAUDE.md. Detects stack, test commands, and conventions. Use when onboarding to a new project or user says 'init', 'setup claude', 'generate CLAUDE.md'.