spike-init

Research a repo, then actually *run* the "how to set up the env / run tests / lint / type-check" commands, and produce a clean pair of agent-context files for the project — `AGENTS.md` (the cross-tool convention every coding agent reads) plus a repo-root `CLAUDE.md` that `@AGENTS.md`-imports it (Claude Code only reads `CLAUDE.md`, not `AGENTS.md`). If those files already exist, *improve* them rather than overwrite — re-research, re-verify, and reconcile each existing claim. Verified commands get a `✅ verified <date>` marker so downstream agents and human reviewers know they were actually executed, not guessed. Use this when the user wants to "set up AGENTS.md / agent context for this project", "this repo has no CLAUDE.md, create one", "the AGENTS.md is out of date, fix it", "bootstrap project context", or "/spike-init". The difference from Claude Code's built-in `/init`: `/init` analyzes the codebase and writes down the commands it *infers*; `/spike-init` runs those commands first and only writes the ones that actually worked.