agentleFS
eduardoweberai-cloudeduardoweberai-cloud/claude-skills/skills/roast/SKILL.md

roast

Skill0 starsChanged 59 days agoMITView on GitHub

Conselho adversarial de 5 personas paralelas + juiz com veredito APROVADO/REMODELADO/MORTO e o teste de 48h mais barato, para stress-testar ideia, plano, decisão, oferta ou trabalho. Use quando o usuário quiser pushback honesto em vez de concordância: /roast, 'critica isso de verdade', 'me dá um roast', 'para de concordar comigo', 'destrói essa ideia', 'isso tem furo?', 'advogado do diabo', 'stress test this', ou quando estiver prestes a se comprometer com um plano e perguntar se sustenta.

32 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use storm-research, to-prd and 7 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • SevenBaylorSkill

    storm-research

    Full STORM multi-agent research pipeline — 5 parallel perspectives, contradiction mapping, independent peer review, cited synthesis. Outputs MD + HTML + PDF.

    Also in 3 projects that use this

  • GoogleSkill

    to-prd

    Turn the current conversation into a PRD and publish it to the project issue tracker — no interview, just synthesis of what you've already discussed.

    Also in 3 projects that use this

  • AWSSkill

    wizard

    Generate an interactive bash wizard that walks a human through steps only they can perform. Use when provisioning infrastructure, setting up credentials or CI secrets, walking an unfamiliar third-party dashboard, or running a one-off migration or cutover. Don't invoke this for steps the agent can perform itself.

    Also in 3 projects that use this

  • engtySkill

    codex

    Use when the user asks to run Codex CLI (codex exec, codex resume) or references OpenAI Codex for code analysis, refactoring, or automated editing

    Also in 3 projects that use this

  • Google CloudSkill

    triage

    Gather context on a change or bug and plan the implementation.

    Also in 4 projects that use this

  • wpankSkill

    session-handoff

    WHAT: Create comprehensive handoff documents that enable fresh AI agents to seamlessly continue work with zero ambiguity. Solves long-running agent context exhaustion problem.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

10 rules. 0 are common, 5 are distinctive, and 5 are very specific to this project.

  • 0

    Common

    Most of the 30 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use whenever the user wants to pressure-test an idea rather than hear approval.

    line 19
  2. Very specific:

    If the user passed it as an argument (/roast <ideia>), use that.

    line 30
  3. Distinctive:

    If the subject is genuinely ambiguous, ask ONE sharp question to pin it down — then proceed.

    line 33
  4. Distinctive:

    Do not interrogate; one question maximum.

    line 34
  5. Very specific:

    Write a tight ## CONTEXT block (3-8 lines) capturing the subject, any constraints the user gave, and what "success" would mean to them.

    line 36
  6. Very specific:

    Read references/council.md.

    line 41
  7. Very specific:

    After all five return, launch ONE more agent (general-purpose) using the JUIZ brief from council.md.

    line 64
  8. Distinctive:

    Keep it direct and unsoftened.

    line 77
  9. Distinctive:

    All persona and judge output is in PT-BR.

    line 82
  10. Distinctive:

    If WebSearch/WebFetch is unavailable in the session, the Pesquisador still runs but must flag that it could not verify externally — never …

    line 83

“Common” is measured: kept by at least 6 in 10 of the 30 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Roast

Purpose

Force a hard, multi-angle critique of whatever the user is proposing, instead of agreeing with it. The skill spins up a council of five adversarial personas in parallel, each attacking the subject from a different angle, then a judge synthesizes everything into a single verdict and one cheap experiment to run next. The point is to break agreement mode: surface fatal flaws, unexamined assumptions, real market data, and the honest reaction of the person who'd be affected — before the user invests time or money.

When to use

Use whenever the user wants to pressure-test an idea rather than hear approval. Triggers include /roast, "critica isso de verdade", "para de concordar comigo", "destrói essa ideia", "isso tem furo?", "advogado do diabo", "stress test this idea", or any moment a plan/offer/ decision is on the table and the user wants to know if it actually survives scrutiny. Works on business ideas, technical plans, content strategy, offers, life/career decisions — anything.

Workflow

1. Identify the subject

Determine exactly what is being roasted:

  • If the user passed it as an argument (/roast <ideia>), use that.
  • Otherwise, use the most substantial idea/plan/decision the user has described in the recent conversation.
  • If the subject is genuinely ambiguous, ask ONE sharp question to pin it down — then proceed. Do not interrogate; one question maximum.

Write a tight ## CONTEXT block (3-8 lines) capturing the subject, any constraints the user gave, and what "success" would mean to them. This block is injected verbatim into every agent.

2. Read the council briefs

Read references/council.md. It contains the five persona prompts and the judge prompt, plus the shared output format. Use those briefs verbatim — they are tuned to force a committed stance from each persona.

3. Dispatch the 5 personas IN PARALLEL

In a single message, launch all five subagents at once (five Agent tool calls together, subagent_type: "general-purpose"). Each agent's prompt = the ## CONTEXT block + that persona's full brief from council.md + the shared output format.

#PersonaAngle
1ContrarianFind the fatal flaw that kills it
2ExpansionistaFind the biggest upside / 10x version
3Primeiros PrincípiosRebuild from pure logic, zero context
4Pesquisador ProfundoPull real web data: market, competitors, precedents, prior art
5Comprador / parte afetadaRole-play the end user — would they actually buy/adopt it?

The Pesquisador must actually use WebSearch/WebFetch — note in its prompt that real sources are required, no invented data.

4. Dispatch the Judge

After all five return, launch ONE more agent (general-purpose) using the JUIZ brief from council.md. Feed it the ## CONTEXT block plus the five persona reports verbatim. Running the judge as a fresh subagent (that never saw the warm conversation) keeps the verdict harsh and independent — do not synthesize it yourself in agreement mode.

5. Present the verdict

Relay to the user, in PT-BR:

  • The judge's VEREDITO (APROVADO / REMODELADO / MORTO) and its reasoning.
  • A condensed digest of each persona's "Tiro mais forte" (one line each) so the user sees the five angles at a glance.
  • The Teste de 48h prominently — this is the actionable payoff.

Keep it direct and unsoftened. Do not append reassurance or hedge the verdict to make the user feel better — that defeats the entire purpose of the skill.

Notes

  • All persona and judge output is in PT-BR. Briefs are written in PT-BR for that reason.
  • If WebSearch/WebFetch is unavailable in the session, the Pesquisador still runs but must flag that it could not verify externally — never let it fabricate numbers.
  • Five-plus-one parallel agents is intentional. Do not collapse it to fewer agents or do it inline; the independence between angles is what produces non-sycophantic critique.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.