ddw-context-checklisted

# Skill: /ddw-context-check ## Description DDW runs commands: the test suite, the linter, the type checker. It learns them from the context file (`AGENTS.md`, `CLAUDE.md`, `GEMINI.md`) — and when that file does not mention one, DDW does not run it. The result is code that passes every gate and then fails your CI, or commits that your `pre-commit` hook rejects for reasons the agent has no way to explain. This skill compares the two: **what the repo already says about how it is built**, against **what the context file told DDW**. It reports the difference and proposes a one-line fix for each. It is **read-only until you say otherwise**, it **never blocks**, and it writes to exactly one file — the context file — and only with your approval. ## What it must NOT report This restriction is the skill, not a caveat on it. **It reports only knowledge the agent cannot discover on its own** — the commands DDW is about to run. It never proposes prose, conventions, architecture, style guidance, or anything the agent can find by reading a file at the moment it needs it. The reason is evidence, not taste. A study across 138 real repositories found that repository context files produced **no improvement in agent task success rate while adding over 20% to inference cost**, with LLM-generated ones actively hurting; the authors' conclusion was that these files should be minimal rather than comprehensive. A skill that grows the context file is therefore not a neutral act — without this r