@rapchic/roast

Evidence-based code roast for AI coding agents — install once, roast any repo with citations. No LLM in the CLI; your agent runs the roast.

Quick start

npx @rapchic/roast install

Restart Cursor, then /roast.

Team repos: commit .cursor/commands/ + .cursor/skills/roast/ so teammates get /roast without a global install.

Global-only: npx @rapchic/roast install --no-project

Slash commands

| Slash | Use when | |-------|----------| | /roast | Roast + fix (daily default) | | /roast-only | Verdict only | | /roast-idea | Critique a plan before code | | /roast-what | Explain the diff or a roast in plain English | | /roast-learn | Learn this project's patterns & antipatterns (once or continuous; optional chat transcripts) | | /roast-install | Reinstall the skill pack | | /roast-full | Optional — same skill loaded directly |

What it does

• Installs a roast skill + slash commands to Cursor, Claude Code, or Codex
• Gathers context (roast context) — stack, scripts, conventions, CI
• Surfaces diff signals (roast diff) — changed files, test/doc hints, suggested scope
• Enforces evidence — findings cite file:line, diff, or test output; chat-only output

Full documentation →

CLI commands

| Command | Description | |---------|-------------| | roast install | Install skill + commands (global + project .cursor/) | | roast update | Update installed skill | | roast status | Version + update check | | roast uninstall | Remove from IDEs | | roast context | Phase 0 INIT — stack, scripts, rules, CI | | roast diff | Git diff signals for roast input | | roast init | Detect repo + IDE, write .roast/config.json | | roast bootstrap | Alias for install (prefer install) |

See docs/commands.md for flags and examples.

Roast modes

| Mode | Trigger | Edits | |------|---------|-------| | roast-only | /roast-only, "roast independently" | No | | roast-and-fix | /roast, "roast and fix" | Yes | | roast-idea | /roast-idea, "roast this idea" | No | | roast-what | /roast-what, "eli5", "what changed" | No | | roast-learn | /roast-learn, "learn this repo" | roast-patterns.mdc only | | roast-then-build | "roast then implement" | After agreement | | roast-then-apply | "preview, roast fix, apply" | After approval | | roast-no-patch | "don't patch", "senior architect" | Structural only |

Details: docs/modes.md

Phase 0 INIT

Before any roast, gather repo context:

npx @rapchic/roast context
npx @rapchic/roast diff --base auto

The agent reads conventions (AGENTS.md, .cursor/rules, CI) and scopes the critique. See docs/init-phase.md.

Examples

Auth middleware

User: /roast-only — review src/middleware/auth.ts

## Roast: auth middleware
**Context:** node — npm · src/middleware/auth.ts · vitest · AGENTS.md
### The real problem
Optional auth treats missing tokens as guest access on routes that require enforcement.
### Findings
- 🔴 Missing enforcement on DELETE — `src/middleware/auth.ts:41` — ...

API handler

User: /roast — review the payment handler

Agent: INIT → findings → minimal fix → npm test

Dashboard component

User: /roast-idea — add infinite scroll to Dashboard

Agent: critiques plan before code — edge cases, a11y, test strategy

More: examples/expected-roast-output.md

Install paths

| IDE | Skill | Commands | Rule | |-----|-------|----------|------| | Cursor | ~/.cursor/skills/roast/ | ~/.cursor/commands/roast*.md | ~/.cursor/rules/roast.mdc | | Claude | ~/.claude/skills/roast/ | — | block in CLAUDE.md | | Codex | ~/.codex/skills/roast/ | — | block in AGENTS.md |

Package name

@rapchic/roast on npm. Bins: roast and roastit.

What roast is not

• Not comedy or profile satire
• Not an LLM API wrapper
• Not a report generator that writes files to your repo

Contributing

See CONTRIBUTING.md. Security: SECURITY.md.

License

MIT — see LICENSE.