dshills

Run plancritic against PLAN.md (or another implementation plan) to validate it before code generation begins. Use when the user invokes /plancritic, asks to review/validate/critique/check a plan, or whenever a PLAN.md is about to be handed off for implementation. Surfaces contradictions, ambiguities, missing prerequisites, risks, test gaps, and ordering problems as a structured JSON critique with a verdict and score.

Use when the user explicitly asks to validate, check, gate, critique, or review a specification (SPEC.md or any *.spec.md / specs/*.md) before planning or implementation. Trigger phrases include "run speccritic", "check the spec", "gate the spec", "validate SPEC.md", "is this spec ready", or similar. Also trigger when the user says they've just finished writing or revising a SPEC and is about to start planning or coding — speccritic is the first gate in the SPEC → PLAN → CODE pipeline and must pass before plancritic runs. Do NOT trigger on incidental mentions of "spec" in unrelated contexts (e.g. test specs, hardware specs, package specs).

Deterministic context planner that runs BEFORE a coding agent starts work. Use this skill whenever a coding task is about to begin against a repository — to score which files the agent should load, fit them into a token budget, detect missing spec/tests/config/runtime context, and compute a feasibility score. Trigger this at the start of any non-trivial coding task, after a plan has been approved, before invoking Claude Code / Codex / any agent on a repo task, whenever the user mentions "let's code this up" or "start implementing" or "run the agent", and any time feasibility of a task is in question. This is a gate, not a coding tool — it decides whether an agent should even start.