Skip to content

heyparsadev/claude-venture-plugin / venture/skills/review

review

Verified

This skill should be used when the user wants an honest overall assessment of a venture — a board-style review, go/no-go decision, kill-pivot-persevere call, investment-readiness check, or periodic check-in — e.g. "review my whole venture", "should I continue with this", "be brutally honest about this business", "monthly business review", "is this worth pursuing", "کل بیزینس رو نقد کن", "بی‌رحمانه بررسی کن", "ادامه بدم یا نه", "جلسه هیئت مدیره", "ارزیابی کلی ونچر". Produces a cross-module scorecard with a kill/pivot/persevere/double-down recommendation and updates the decision log.

Venture Review — The Board Meeting

Step back from module-level work and answer the only question that matters: is this venture working, and what's the highest-leverage next move? Like a good board: evidence-driven, kind to the founder, merciless to the plan.

Shared conventions (venture suite)

  • Workspace: locate ventures/ in the current working directory (ask once and create if missing). All outputs go to ventures/<slug>/<NN-module>/.
  • Templates are read-only. Files under a skill's references/ live inside the installed plugin and are shared by every venture. Never write to them. "Fill template X" always means: copy it into ventures/<slug>/<NN-module>/ under a dated name and fill the copy.
  • Read VENTURE.md first. When done: update its Status table and append key findings to "Key facts & numbers" (format: - [module, YYYY-MM-DD] finding (source, confidence)).
  • Language: converse in the user's language. Write formal deliverables (reports, canvases, battlecards) in the report_language set in VENTURE.md. Persian reports keep standard English business terms as-is (TAM, SAM, SOM, CAC, LTV, GTM, churn, retention...).
  • Research integrity: cite every external claim (source + date). Mark every estimate with confidence (high/med/low). Never fabricate numbers — write "no data found" instead.
  • Chain: end by updating VENTURE.md and suggesting the logical next module.

Workflow

1. Read the whole venture

Everything: VENTURE.md, all module reports (latest dated file per module), decision log, open assumptions. Plus ask the user: what changed since the last review? (new numbers, customer events, money/runway, morale — runway and energy are real variables; ask, respectfully).

2. Cross-module scorecard

DimensionSourceGrade (A–F)EvidenceFreshness
Problem & demand validation01-discovery
Market attractiveness02-market
Competitive position03-competitors
Product-market fit04-pmf
GTM execution & traction05-gtm + user data
Economics viability06-model

Rules: a grade needs evidence from the files — no evidence, grade is ? (and that's a finding: "we don't know X" is a top risk). Mark stale anything >90 days old or invalidated by newer findings (e.g. model built on a price GTM later changed) and set those modules stale in VENTURE.md.

3. Coherence check (the cross-module value — modules can each pass while the whole fails)

  • Does the GTM target segment = the segment discovery/PMF found actually loves it?
  • Does pricing in the model = pricing in GTM = willingness-to-pay in discovery?
  • Does SOM support the model's growth assumption?
  • Did any module's "riskiest assumption" get quietly forgotten? List the still-untested ones with age.

4. Adversarial pass — mandatory

Launch venture:devils-advocate with: venture one-liner, the scorecard + your tentative recommendation, paths to the 2–3 most load-bearing reports, and "what decision rides on this" (continue/kill/raise/invest more time). Integrate: address every critical objection — accept (changes the recommendation), rebut (with evidence, cited), or downgrade confidence visibly.

5. Verdict

<slug>/venture-review-YYYY-MM-DD.md (in report_language):

  1. State of the venture — one honest paragraph.
  2. Scorecard (above) + coherence findings.
  3. Top 3 risks, ranked, each with cheapest mitigation/test.
  4. Recommendation — pick one, commit:
    • Double-down — evidence of working engine; name what to pour fuel on and what to stop doing (focus = stopping things).
    • Persevere — on track, no step-change; name the milestone and date for next review.
    • Pivot — core assumption broke but an asset is real (a retaining segment, a channel that converts, a team skill); name 1–2 concrete pivot directions grounded in the strongest evidence found, and the module sequence to validate them.
    • Kill — kindly, plainly, with the evidence; what was learned, what's salvageable (relationships, code, audience). A clean kill is a win, not a failure — say that and mean it.
  5. Next 30 days — 3–5 actions max, each mapped to a module or real-world task.
  6. For client ventures: add a client-facing executive summary (1 page, diplomatic but unspun).

6. Close the loop

Append the recommendation + rationale to VENTURE.md Decision log (date, decision, rationale — one row). Update statuses/staleness. Set the next review trigger ("after 30 days or when <metric/event>").

Quality bar

  • The recommendation must be falsifiable: "persevere IF X by DATE" — never "keep going and see".
  • Sunk cost is not evidence; "we've worked on it 8 months" appears nowhere in the reasoning.
  • Respect the human: deliver hard verdicts with care, but never soften them into ambiguity — ambiguity steals months.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 470kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall