do-roast
Roasts code in a character voice, each joke tied to a real finding. Use when the user wants a funny review of code ("roast this code", "do-roast", "make fun of", "character roast of"). Not for a serio
SKILL.mdplugins/mz-funny/skills/do-roast/SKILL.md
GitHubEvidence-Anchored Character Roasting Pipeline
Overview
You orchestrate a character-voice code roast. Static analysis, docs coherence, and web best-practice research produce a structured dossier of numbered findings; the user's chosen persona agent renders a roast that may embellish tone but may not invent substance. Every roast line must cite a real finding.
When to Use
A funny roast, in a chosen persona's voice, of code, a branch, a PR or text, where every roast line cites a real finding.
When NOT to use
- Actual code review — use
/review-branchor/audit. - Actual planning — use
/build. - Serious architecture critique — use
/explain. - The target does not exist or cannot be resolved to concrete files or text.
Input
$ARGUMENTS is split on first whitespace. Token 1 = persona (validated against PERSONA_ALLOWLIST, case-insensitive). Remainder = free-form target prompt. Empty $ARGUMENTS → ask. Unknown persona → ask with allowlist. Empty remainder → ask.
Constants
- PERSONA_ALLOWLIST:
["caveman", "wh40k-ork", "pirate", "viking", "dwarf", "drill-sergeant", "yoda"](lowercase canonical) - MAX_PERSONAS: 7 (allowlist bound; single-persona dispatch per invocation)
- PROJECT_ROOT: resolved once in setup with
echo "${MZ_MEMORY_ROOT:-$(git rev-parse --show-toplevel 2>/dev/null || pwd)}"; expand<PROJECT_ROOT>to it in every command and agent prompt. - TASK_DIR:
<PROJECT_ROOT>/.mz/task/ - REPORT_DIR:
<PROJECT_ROOT>/.mz/reports/ - RESEARCH_CACHE_DIR:
<PROJECT_ROOT>/.mz/research/(stack-wide web cache, 7-day staleness) - MAX_FINDINGS: 30 (upper bound on dossier entries per invocation)
- Note:
MAX_FINDINGS=30is the upper bound. Analysis should prefer 10-15 high-quality Critical/Nit findings over 30 weak Optional/FYI — reduces dispatch token cost and improves voice focus.
- Note:
- DOSSIER_SEVERITY_LABELS:
["Critical", "Nit", "Optional", "FYI"]
Available Personas
<!-- verify: every value in PERSONA_ALLOWLIST must appear in this table -->| Persona | Voice | Best for roasting |
|---|---|---|
| caveman | Hulk-speak, pre-linguistic | Over-engineered abstractions — "me no understand, me smash" |
| wh40k-ork | Greenskin dakka warrior | Weedy/overclever code ("dat code iz weedy") |
| pirate | Robert-Newton West-Country | Ships that won't sail — bloated files, leaks, "Davy Jones" |
| viking | Skald with kennings | Forgettable code, no sagas, "argr" / ignoble logic |
| dwarf | Scottish brogue + Dammaz Kron | Shoddy craftsmanship — "umgak", grudge book additions |
| drill-sergeant | FMJ Gunnery-Sergeant cadence | Discipline failures, fitness metaphors, profanity allowed |
| yoda | OSV-inverted disappointment | Clouded logic, "much to learn" disappointment register |
Core Process
Phase Overview
| # | Phase | Details |
|---|---|---|
| 0 | Setup + arg parse | Inline below |
| 0.5 | Target resolution | Inline below |
| 1 | Structural/smell analysis | phases/analyze.md |
| 2 | Docs coherence analysis | phases/analyze.md |
| 3 | Web best-practice research | phases/analyze.md |
| 4 | Dossier writer | phases/analyze.md |
| 5 | Persona dispatch | phases/render.md |
| 6 | Report assembly + teaser | phases/render.md |
Common Rationalizations
| Rationalization | Rebuttal |
|---|---|
| "Just this once, the persona can riff on something not in the dossier — it's funnier." | The whole plugin's value is evidence anchoring. One fabricated line poisons every future roast's credibility. Cut it. |
| "The code probably has this bug even though analysis didn't find it — persona can mention it." | If analysis missed it, analysis gets re-run. The persona may not invent findings. Re-run Phase 1-4, do not extrapolate. |
| "The user will never notice a made-up line if the rest is grounded." | The user will grep the roast for Finding N citations. Fabrication is detectable and contractually forbidden. |
Red Flags
- Dossier contains fewer than 3 findings but persona roast is >200 lines — embellishment exceeded substance.
- Roast mentions a file path that does not appear in the dossier — fabrication.
- Persona agent was dispatched without the dossier inlined in the prompt.
- Any author name, git blame output, or real-person reference in the dossier or report.
- A natural-language target with several candidates was roasted without asking which ones.
Verification
Output a visible block showing: task name, resolved target, persona chosen, finding count by severity, report path, and the rendered teaser paragraph. Confirm the report file exists and is non-empty. Confirm every Finding N in the dossier has at least one reference in the rendered roast (soft check).
Phase 0: Setup
Resolve PROJECT_ROOT with the command in Constants, before any cd or worktree switch, and use it for every .mz/ path below. Parse $ARGUMENTS: split on first whitespace; validate token 1 against PERSONA_ALLOWLIST (lowercase compare). Unknown → AskUserQuestion listing allowed personas. Empty remainder → AskUserQuestion asking what to roast. Skill name do-roast → task dir prefix do_roast (snake_case). Derive task name: <YYYY_MM_DD>_do_roast_<slug> where slug is a snake_case summary (max 20 chars) of the resolved target or first 3 words of the remainder, and <YYYY_MM_DD> is today's date with underscores. Create TASK_DIR<task_name>/. Write state.md with schema_version: 2 as the first line, then Status: running, Phase: setup, Started: <ISO8601>, phase_complete: false, what_remains: [], Persona: <chosen>, Target_raw: <remainder>.
Phase 0.5: Target Resolution
Read phases/analyze.md ("Phase 0.5: Target resolution detail"), run its ladder on Target_raw, and record the result in state.md as that section describes. Explicit paths, a branch diff and raw text name their target directly, so the roast goes on to Phase 1 without asking.
Only a natural-language target (rung 4) that leaves more than one candidate needs the user. Ask with an input collector, not an approval gate: no plan message and no Approve/Reject.
- Set
what_remains: [awaiting answer: target], keepStatus: running, and call AskUserQuestion once: questionWhich files should the <persona> roast for "<Target_raw>"?, headerTarget,multiSelect: true, one option per candidate with the path as the label and why it matched as the description. - Record the chosen paths as
resolved_target_list, setwhat_remains: [], and go on to Phase 1. Run a typedOtheranswer through the ladder again. - If nobody can answer, set
Status: blockedand stop.
Error Handling
- Empty
$ARGUMENTS→ AskUserQuestion (never guess). - Unknown persona → AskUserQuestion with
PERSONA_ALLOWLIST. - Target resolves to 0 files AND is not raw text → escalate with
STATUS: BLOCKEDand prompt user. - Persona agent returns empty → retry once with clarified prompt; still empty → set
Status: completewith the concern under## Notesinstate.md, and include the raw dossier in the report.
State Management
Update TASK_DIR<task_name>/state.md after every phase transition. Track: current phase, persona, resolved target list, finding counts by DOSSIER_SEVERITY_LABELS, report path.
State persists to .mz/task/<task_name>/state.md under the project root, ${MZ_MEMORY_ROOT:-$(git rev-parse --show-toplevel 2>/dev/null || pwd)}, resolved once at the start of the run. Schema is v2: line 1 is schema_version: 2; line 2 is Status: with one of pending, running, blocked, complete, aborted_by_user or failed, lowercase and unquoted; then Phase, Started, phase_complete (boolean) and what_remains (YAML list of strings), with skill-specific keys below them. Milestones go in Phase; the reason for blocked or failed goes in what_remains. Set phase_complete: false on phase entry and true once the phase's artifacts are written and its gates pass. Refresh what_remains on every phase transition; it is [] when Status is complete. Write the file after every phase and before every gate, and re-read it after compaction. Record decisions worth recalling under a heading that reads exactly ## Decisions, placed before ## Notes; mz-memory harvests them only once Status is complete. Under a heading that reads exactly ## Lessons, add one line whenever a review finding is accepted and fixed, a step fails, gate feedback corrects a mistake, or an approved approach is reversed, in the form - (review|failure|correction|change) WHEN <situation> → <what to do> — WHY <cause>, concrete and under about 40 words; mz-memory harvests lessons whatever the Status. Treat a legacy completed as complete. On reading a schema_version: 1 or unversioned file, put schema_version: 2 on line 1 and Status on line 2, add the missing keys, and log the upgrade. Leave a file with a higher schema_version untouched: ask whether to Restart (archive it), start New (_v2) or Abort, and stop if nobody can answer.
Critical: every line of the rendered roast must trace to a numbered Finding in the dossier. No invention. No real-person attacks. No git blame. If the persona cannot find something to roast within the evidence, it says so — it does not fabricate.