Agent-Skills.md

Agent Skills: Code Roast

>

UncategorizedID: likw99/agent-skills/code-roast

Repository

likw99/agent-skills
likw99License: MIT

Install this agent skill to your local

pnpm dlx add-skill https://github.com/likw99/agent-skills/tree/HEAD/skills/code-roast

Skill Files

Browse the full folder contents for code-roast.

Download Skill

Loading file tree…

skills/code-roast/SKILL.md

Skill Metadata

Name
code-roast
Description
>

Code Roast

Give any codebase a Gordon-Ramsay-style roast: a letter grade, a Hall of Shame, genuine Bright Spots, and a Prescription — all in a shareable Markdown report.

Workflow

Step 1 — Identify the repo root

If the user didn't specify a path, use the current working directory as the repo root. Confirm it is a git repo or a directory with code files. Store it as <repo_root>.

Step 2 — Run the analyzer

cd <skill_dir>/scripts
uv run analyze.py <repo_root> --debug

This emits a JSON object to stdout with 9 shame-category metrics. Copy the full JSON — you'll need it in Step 3.

If uv is not available, fall back to: python3 analyze.py <repo_root> --debug

Step 3 — Score and write the roast

  1. Open references/roast-rubric.md and follow it exactly:

    • Apply the penalty table for each of the 9 categories to get a total penalty score.
    • Look up the grade tier and its tagline.
    • Write each section using the section templates as your guide.

  2. Write the roast to <repo_root>/code_roast_YYYY-MM-DD.md using today's date.

  3. The report must contain exactly these sections in order:

    • Header with repo name, grade, tagline, and file/line counts
    • The Verdict — 2–3 sentence overall tone-setter
    • Hall of Shame — one subsection per triggered category (penalty > 0), skipping clean ones
    • Bright Spots — 1–3 genuine positives
    • The Prescription — 3–5 numbered, actionable fixes
    • Footer with the shareable X/Twitter link (pre-filled with the grade)

Step 4 — Present the roast

After writing the file, display the full roast inline in the conversation so the user sees it immediately. Then tell them the file path.

Key Rules

  • Be specific: always use real file names, real counts, real commit messages. Vagueness is not funny.
  • Skip clean categories: if a category has penalty 0, don't mention it in the Hall of Shame.
  • Always end with empathy: The Prescription and Bright Spots soften the roast. Never end on pure shame.
  • Tone: dry wit, not cruelty. The rubric's voice section has examples.
  • Length: 300–600 lines in the output file is ideal — long enough to be thorough, short enough to share.

Resources

  • scripts/analyze.py — static analysis engine; outputs JSON metrics to stdout
  • references/roast-rubric.md — scoring tables, grade tiers, tone guide, section templates, output file spec