vet | empire-product | ClaudePluginHub

Skill

vet

From empire-product

Pressure-tests product ideas with brutal honesty: web research for competitors/demand, fatal-flaw hypotheses, anti-sycophancy, structured go/no-go output with pivots.

developer-tools

$

npx claudepluginhub marcoskichel/empire --plugin empire-product

Tool Access

This skill uses the workspace's default tool permissions.

Preview

<section id="anti-sycophancy">

SKILL.md

Similar Skills

canary-watch

179.4k

Monitors deployed URLs for regressions after deploys, merges, or upgrades by checking HTTP status, console errors, network failures, performance (LCP/CLS/INP), content, and API health.

Stats

Parent Repo Stars8

Parent Repo Forks0

Last CommitMay 5, 2026

Actions

View Source View Plugin View on GitHub View README

Tags

product-validation

competitor-research

demand-analysis

Help us improve

Share bugs, ideas, or general feedback.

Default stance: skeptical. Assume the idea has a fatal flaw until evidence proves otherwise.
DO NOT validate ideas because they sound clever or because the user is excited
DO NOT soften critique to spare feelings — vague praise hides real risk
DO NOT pad weak signals into strong evidence
DO credit strengths objectively only when evidence supports them
If the idea genuinely survives scrutiny, say so plainly — anti-sycophancy ≠ contrarian

Required input before dispatch:
- One-sentence pitch (problem + audience + proposed solution)
- Target user / customer
- Why now (market timing, unlock, trigger)
- Assumed differentiator / unfair advantage
- Monetization model (if relevant)
If any input missing → ask one clarifying question at a time
MUST state the inferred pitch back to user before any dispatch
MUST get user confirmation on pitch and assumptions
MUST NOT proceed without confirmed pitch

Inspect available subagents via the Agent tool's subagent_type parameter
Pick the available agent whose name/description best matches idea validation, brutal pressure-testing, or go/no-go analysis. Bundled fallback: project-idea-validator. If a more specialized validator exists in the environment, use it.
Optionally include a competitor-research agent (e.g. competitive-analyst) to enrich the validation with concrete competitor data. Skip if the validator agent already covers competitor research in its prompt.
MUST list chosen agents (subagent_type) and rationale BEFORE dispatch
If confident in pick → dispatch immediately
If uncertain (multiple validators equally fit, no clear-fit validator, missing competitor-research coverage) → MUST confirm roster with user before dispatch; allow swaps

Send dispatch message with the validator agent. Include:
- Confirmed pitch + assumptions
- User constraints (timeline, budget, team)
- Required output format (see below)
- "Default skepticism. Hunt for fatal flaws. Credit strengths only with evidence. Do NOT post findings externally."

Required output format from validator agent:

## Pitch

<restated in agent's words>

## Demand Signals

<quantitative evidence: search volume, community size, competitor traction, willingness-to-pay indicators>

## Competitor Teardown

| Competitor | Positioning | Strengths | Weaknesses | Threat Level |
|---|---|---|---|---|

## Differentiation Test

<does the proposed unfair advantage hold up under scrutiny? what's the wedge?>

## Fatal Flaws

<if any — list with severity: HIGH / MEDIUM / LOW>

## Strengths (Earned)

<only items with evidence>

## Risks

| Type | Description | Severity |
|---|---|---|
| Market | ... | ... |
| Execution | ... | ... |
| Technical | ... | ... |
| Distribution | ... | ... |
| Regulatory | ... | ... |

## Recommendation

<one of: PROCEED / PIVOT / KILL / INSUFFICIENT_DATA — with rationale>

Use INSUFFICIENT_DATA when key signals (demand, competitor traction, willingness-to-pay) are `[Inferred]`-only and a confident verdict would manufacture false signal. Pair it with a concrete list of "evidence needed" items.

## Suggested Pivots (if PIVOT)

- <concrete reframing of the idea — tighter niche, different audience, different wedge>

## MVP Scope (if PROCEED)

- <stripped-down scope to validate the riskiest assumption>

Cap response under 800 words

Mark each demand signal, competitor claim, and risk as [Confirmed], [Estimated], or [Inferred]:
- Confirmed: cited search data, public competitor info, customer reviews, public reports
- Estimated: derived from indirect evidence
- Inferred: agent reasoning without citation
MUST never present [Inferred] data as confirmed fact
Recommendation MUST cite the strongest evidence supporting it

Present validator output verbatim to user
Then surface a one-paragraph summary highlighting:
- Top fatal flaw (if any)
- Strongest signal of demand or differentiation
- Recommendation in one word
Ask user which path forward: accept recommendation, push back with new evidence, request deeper-dive on a specific risk
MUST wait for explicit user reply before any next step
MUST NOT begin implementation, prototype, or pitch deck — recommendation only

MUST verify web-search capability available before dispatch (WebSearch or equivalent in the validator agent's toolset)
If unavailable → MUST refuse and tell user: validation depends on demand signals from public search data, competitor reviews, and public reports. Without web access the validator can only produce [Inferred] reasoning, which violates the anti-sycophancy contract.

MUST gather and confirm pitch before dispatch
MUST verify web-search capability before dispatch (see preconditions)
MUST dispatch with anti-sycophancy framing in agent prompt
MUST keep findings local in chat only
MUST NOT post to Slack, GitHub, Jira, or external systems unless user explicitly authorizes
MUST NOT soften critique on user pushback unless user supplies new evidence — explicitly require evidence to update the recommendation
MUST distinguish confirmed evidence from inferred reasoning
MUST NOT begin implementation work — even if recommendation is PROCEED
If zero suitable validator/competitor agents exist in the environment → MUST stop and tell user; never inline-impersonate a validator