left with it; CI covers it through `scripts/check-mux-names.test.ts`.
- **A finding is fixed in the code, never suppressed and never cleared by downgrading a rule.**
There are zero `oxlint-disable` comments
frontend) or `packages/app/server/i18n.js` (server/CLI) — adding them to the wrong file is prohibited.
After any code change, `pnpm run build` must be run.
Never commit `package-lock.json`. Any new dependency that ships
sight. SKILL.md stays under 500 lines with header AND footer version bumped together.
- Harness code invariants (see `docs/dev/architecture-reference.md`): new always-on prompt text goes in the cache-stable prefix; `goal_score.ts
breathe. The project brief controls pacing, palette, and typography.
Framework skill contracts override historical code recipes in the style guide.
## Workspace
Create a project with `npm run new-video
watcher tracks CI checks, review status, and merge state
7. Auto-triggers codereview agent on CI pass or PR open (if enabled)
8. Auto-resumes agent when reviewer requests
drift and replay controls, not proof of human review.
Mark paid MCP tools with `anthropic/requiresUserInteraction`; Claude Code
2.1.199+ forces an interactive host decision even under allow, Auto, or
Bypass modes
proven implementation patterns in `references/` for Claude to adapt.
- Follows the official Claude Code plugin spec
## Directory Structure
```
claude-skills/
├── plugins/ # one folder per plugin; the folders are the truth
audit script: `.claude/skills/file-headers/` (verify with `bash .claude/skills/file-headers/scripts/check-headers.sh`). This binds every coding agent (Claude Code, Codex, or others).
## Commands you should know
- Setup: `npm run setup`
- Dev: `npm run dev`
- Prod
code, read exports, immediate callers, shared utilities.
"Looks orthogonal" is dangerous. If unsure why code is structured a way, ask.
### Rule 9 — Tests verify intent, not just behavior
Tests must
Claude Code Instructions
@../AGENTS.md
Follow the imported agent-specific repository policy for all code, review,
commit, and pull request work. Consult the linked human process documents on
demand only when
code
./gradlew check # Run all checks
# Client-side
cd client
npm run format # Format code
npm run check # Run lint, typecheck, and tests
```
### Creating Custom Components via CLI
ByteChef includes
tries on a provider |
**Harness and runtime**
| Use | Never write | Means |
|---|---|---|
| harness | host | Claude Code driving the session; "host" is the MCP host that spawns the server |
| foreign model | external
rewriting
it from memory. The harness CLI (`autonomous-harness`) still carries its own grid code
(`gridLaunch.ts`, `gridWebMcp.ts`, `agent_retarget`, `harness grid login`); nothing here calls it.
## Toolchain and commands
`pubspec.yaml
subagents. Five domain experts that can advise OR implement: Architect, Plan Reviewer, Scope Analyst, CodeReviewer, and Security Analyst.
## Development Commands
```bash
# Test plugin locally (loads from working directory)
claude
dispatch multiple subagents in parallel. Don't serialize work that doesn't need serializing.
### Codereview
When the user asks for a codereview or reviewer, read `.claude/agents/review.md` first
A file Claude Code reads at the start of every session. It holds the commands, conventions and warnings the agent needs for this project.
Where does it go?
At the repository root. Claude Code also reads CLAUDE.md files in subdirectories when it works there.
What should it contain?
Build and test commands, the project's layout, conventions that aren't obvious from the code, and mistakes to avoid. Short files tend to work better than long ones.
CLAUDE.md or AGENTS.md?
Claude Code reads CLAUDE.md; most other agents read AGENTS.md. Many projects keep one and point the other at it.