startup-evidence
Skill9 starsChanged 17 days agoMITView on GitHub
Create evidence-grounded startup strategy and investor materials with claim provenance and an independent verification pass. Use for competitor deep dives, pricing and feature comparisons, battle cards, April Dunford-style positioning, investor IR, 30-day plans, or customer-interview validation. Do not use for generic business prose that needs no research or for live outreach, fundraising, publishing, or account actions.
In this repository56 other files it gives its agents.
- harness/CLAUDE.mdCLAUDE.md
- academic-research-routerSkill
- accessibilitySkill
- adaptive-project-harnessSkill
- anti-slopSkill
- apple-designSkill
- arenaSkill
- astryx-design-systemSkill
- attention-spanSkill
- copywritingSkill
- crash-outSkill
- croSkill
- cybersecurity-skill-routerSkill
- design-isms-referenceSkill
- design-minimal-blue-pitchdeckSkill
- design-source-routerSkill
- design-specSkill
- diagnosing-bugsSkill
- emil-design-engSkill
- external-review-routerSkill
- frontend-designSkill
- gh-address-commentsSkill
- handoffSkill
- humanize-japaneseSkill
- hyperframes-animationSkill
- hyperframes-audioSkill
- hyperframes-cliSkill
- hyperframes-coreSkill
- hyperframes-creativeSkill
- hyperframes-keyframesSkill
- hyperframes-registrySkill
- hyperframesSkill
- hyperframes-studioSkill
- impeccableSkill
- implement-changeSkill
- leak-checkSkill
- library-docsSkill
- loop-librarySkill
- media-useSkill
- motion-graphicsSkill
- obsidian-memory-routerSkill
- openpencil-cliSkill
- ponytailSkill
- product-marketingSkill
- project-startSkill
- pr-ossSkill
- remove-ai-marksSkill
- requirements-clarificationSkill
- retroSkill
- review-changeSkill
- screenshotSkill
- security-auditSkill
- taste-guided-uiSkill
- team-patternsSkill
- understandSkill
- verification-before-completionSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 46
If no independent context is available, report self-check only.
- Distinctive: line 47
Run deterministic checks before judgment: arithmetic, units, currency and billing normalization, repeated-number consistency, required …
- Distinctive: line 48
Use project-native architecture or policy checks when the deliverable includes implementation claims.
- Distinctive: line 54
Never invent competitor prices or features, customer quotes, traction, market size, conversion, revenue, partnerships, or investor interest.
- Very specific: line 55
Treat contact sales, missing plan details, roadmap items, beta features, and inferred capabilities as distinct states.
- Distinctive: line 56
Do not bypass authentication, scrape against access controls, upload private documents, contact customers or investors, publish, update a …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Startup Evidence
Produce decision-ready startup work without turning guesses into facts.
Route the request
- Identify only the requested deliverable modes:
- competitor deep dive, price/feature comparison, or battle card: read competitive-intelligence.md
- positioning, 30-day validation plan, or customer interviews: read positioning-validation.md
- investor memo, IR deck, or fundraising narrative: read investor-ir.md
- verification-only request: do not rewrite the artifact unless the user asks for corrections
- Always read independent-verification.md.
- Establish the decision frame before research: decision, audience, company and product stage, geography, currency, comparison set, artifact format, and an explicit
as ofdate. Infer only low-risk details and label them. - For claims that can change, research current primary sources. Prefer official pricing, product documentation, regulatory filings, customer-authored evidence, and first-party announcements. Use secondary sources only to locate or triangulate primary evidence.
- Build one shared claim ledger and reuse it across every requested artifact. Do not run separate research passes that silently use different scopes or dates.
- Draft the requested deliverable, then run the independent verification gate. One correction and recheck is the normal limit; further rounds require a bounded, approved Goal/UW loop.
Keep a claim ledger
Give every decision-relevant claim a stable ID. Record:
| Field | Required meaning |
claim_id | Stable identifier used by the deliverable and review report |
claim | Atomic statement; split compound claims |
kind | observed_fact, inference, recommendation, estimate, forecast, target, or unknown |
source | Direct URL, local evidence path, or not publicly available |
observed_at | Date the source was checked |
scope | Plan, region, currency, segment, version, period, or other qualifier |
confidence | high, medium, or low, with a short reason when not high |
status | Latest verifier verdict |
supporting_claim_ids | Required for an inference; IDs of the facts or prior inferences it depends on |
reasoning | Required for an inference; concise explanation of how the supporting claims imply it |
decision_criterion | Required for a recommendation; the named objective or threshold it optimizes |
Facts need evidence. Inferences need both supporting claim IDs and reasoning. Recommendations need the decision criterion they optimize. Estimates, forecasts, and targets must never be rendered as actual results. Preserve unknown instead of filling a gap with plausible prose.
Apply the independent gate
- Separate producer and verifier contexts. Prefer a fresh read-only subagent that receives the frozen request, artifact, sources, ledger, formulas, and rubric—not the producer's intended verdict or suspected defects.
- If no independent context is available, report
self-check only; do not call it independent review. - Run deterministic checks before judgment: arithmetic, units, currency and billing normalization, repeated-number consistency, required fields, links, citations, and artifact parsing/rendering.
- Use project-native architecture or policy checks when the deliverable includes implementation claims. Add tools such as dependency-cruiser only when the target is appropriate, the tool is already governed or preflighted, and the user authorized any installation. AI agreement is not an architecture test.
- External AI review is optional and advisory. Route it through the governing external-review workflow; never treat invocation as permission to upload private material.
- A verifier may return
PASS,PASS_WITH_NOTES,FAIL, orBLOCKED. Do not suppress disagreement or uncertainty to make the result look complete.
Respect boundaries
- Never invent competitor prices or features, customer quotes, traction, market size, conversion, revenue, partnerships, or investor interest.
- Treat
contact sales, missing plan details, roadmap items, beta features, and inferred capabilities as distinct states. - Do not bypass authentication, scrape against access controls, upload private documents, contact customers or investors, publish, update a CRM, spend money, or make financial commitments without explicit authorization through a safe tool path.
- Forecasts and fundraising materials are decision support, not financial or legal advice. Surface material assumptions and recommend qualified review where stakes require it.
- Route
.docxIR work through the document skill and presentation/Slides work through the presentation skill so the final artifact is rendered and visually checked. - Store only sanitized, reusable decisions or workflow lessons in memory. Never store raw interviews, investor lists, private decks, credentials, or full source packets.
Return a complete handoff
Return, in the user's requested language and format:
- the requested deliverable;
- a compact claim index or ledger link;
- the independent verification report and independence status;
- explicit gaps, contradictions, and stale evidence;
- the
as ofdate, checks run, and any skipped checks.
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
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.

