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.
2026-04-19