Goal Prompts Get started
← goal prompts · dogfood

Sample reports.

The Day-1 New Repo playbook, run against this catalog's own repository. Every finding cites real code; two of the bugs it found were fixed in the same release. This is the output you can expect from a brief — and the fix log shows the loop closing, where those reports became commits.

00Product Improvement Discovery

5 quick wins and 3 big bets — including a real scoring bug the MCP smoke test surfaced.

→ IMPROVEMENTS.md

01Bug Hunt

5 findings with triggers and root causes; the two S2s (a label race and a self-XSS) shipped fixed in 0.4.

→ BUGS.md

06Security & Privacy Audit

An honest posture summary — and the finding that gave the installer its checksum verification.

→ SECURITY-AUDIT.md

14New-Dev Onboarding Audit

20–30 minutes to a first PR, three real stumbles, and the one-hour fix that removes them.

→ DX.md

47The Fixer

The acting half: the findings above became eight commits in the 0.5 cycle, one per finding, each verified. This is that log.

→ FIXLOG.md

How to read one of these: every finding carries a severity (S1 data loss / security · S2 broken for some users · S3 annoyance) and a confidence (certain / likely / worth-verifying). Findings cite file and line, so you can open the code and check the claim yourself — start with the verification plan for anything marked worth-verifying, and treat a report as testimony, not verdict: the agent argues, the evidence decides.

Day-1, on a repo that isn't this one

The obvious objection to self-dogfood: "of course it works on the repo it was tuned on." So the same Day-1 briefs were run — by coding agents, unattended — against agentic-dev-os, the maintainer's agent-governance repo. Same method, foreign codebase, reports committed unedited.

00Product Improvement Discovery

15 evidenced items, every one empirically reproduced — led by a governance loop whose dry-run can fabricate passing verification events, and a demo that deletes the real event history.

→ examples/agentic-dev-os/IMPROVEMENTS.md

01Bug Hunt

10 findings — 2 S1, 6 S2, 2 S3. The scariest: the repo's own make demo, listed as a ticket verification step, irreversibly wipes the real event log. A second, independent session re-verified every finding.

→ examples/agentic-dev-os/BUGS.md

14New-Dev Onboarding Audit

Setup is flawless — verify green in 0.8 s — but the first PR takes 3.5–5 hours of undocumented process traps; 12 staged fixes, and the one-hour change that removes most of them.

→ examples/agentic-dev-os/DX.md

The security audit (06) ran too — and it's deliberately not published: its top finding is an unpatched command-execution path, so the report went to the maintainer privately until the fix ships. That's the house style working: evidence rules apply to marketing, and ask-first applies to disclosure.

Venture · Gut Check

A second dogfood: the Venture family run before a product exists. Briefs 62 → 63 → 67 research one candidate niche — flaky-test tooling for CI — entirely from sourced web evidence, and rule on it. Every claim carries a link and an access date; the bear case gets equal weight; the verdict comes out pivot, not a reflexive go.

62Pain & Demand Mining

Is the pain real? Vendor headlines, named practitioners, and Google's own "we can't fully count the cost" — graded, with the disconfirming read given equal column inches.

→ examples/venture/DEMAND.md

63Competitor Teardown

Everyone fighting for the budget — Trunk, BuildPulse, Datadog, and free retry plugins — with the gaps classified: open-because-hard vs open-for-a-reason.

→ examples/venture/COMPETITORS.md

67Venture Verdict

Bars set before scoring, then the ruling: pivot — the pain is real but detection is commoditizing, so the wedge moves from "quarantine" to "root-cause & prevent."

→ examples/venture/VERDICT.md

Growth · the whole funnel, audited

The freshest dogfood: all eleven Growth briefs, run against this site as one conductor pass. They are deliberately self-critical — the competitive scan records the 0-star gap in the open, and the proof audit called out the site's own overclaims. Their findings became the roadmap the site is shipping right now.

09Funnel Friction Audit

The real funnel reconstructed — no signup, no backend — with every silent stall between landing and a first report, counted and ranked.

→ FUNNEL.md

10Competitive Gap Scan

Five rivals at 23k–50k stars, table stakes vs differentiation — and the honest headline that the category gap here is distribution and proof, not features.

→ COMPETITIVE.md

11Monetization Map

Where money could enter a free, MIT catalog — partner playbooks, sponsorship rails — and the billing friction standing in the way.

→ REVENUE.md

12AI Opportunity Scan

Which LLM features would earn their place on a no-backend static site — and which would break the "nothing leaves your machine" stance that is the moat.

→ AI-IDEAS.md

70SEO & Discoverability

Titles, structured data, crawlability, and unfurls across 180+ generated pages — including the briefs invisible to non-rendering crawlers.

→ SEO.md

75Conversion Rate Optimization

Every conversion surface for a first-time visitor — including the install path that dead-ended after step 1 and the 176 side-door pages missing the offer.

→ CRO.md

78Retention & Lifecycle Audit

What earns a second visit: the run tracker, the weekly Vitals ritual, the welcome-back context — and the churn cliffs that quietly lose it.

→ RETENTION.md

79Social Proof & Credibility

The site's own trust claims stress-tested — it caught the "every brief is dogfooded" overclaim and an unfalsifiable commit citation; both are fixed in the release that shipped this section.

→ PROOF.md

80Activation & First-Win Audit

The first session, cold arrival to first report file, with every handoff where a newcomer stalls between copy and paste.

→ ACTIVATION.md

109Forms & Validation Audit

No classic forms here — so the audit covers what exists: search, the 4-field context box, file drops, and how kindly each one fails.

→ FORMS.md

110Checkout & Payment Flow

A null report, filed honestly: there is no checkout to audit on a $0 site — the seven lenses had nothing to attach to, and saying so is the finding.

→ CHECKOUT.md

Craft · does it feel right?

The newest family, run on this site the day it shipped. Every finding was reproduced in a real browser — including the one-click removal in Report Studio that silently destroys checklist progress, caught by the family's own first lens: undo over confirm.

152Fit & Finish Audit

The seams after forty logged fix rounds: the × that eats an afternoon of triage, the paste draft that dies with the tab, Escape meaning different things on different pages — and the undo primitive the codebase still doesn't have.

→ POLISH.md

Reproduce this: paste the Day-1 conductor (/raw/playbook-day1.md) into your agent inside any repo — it fetches briefs 00 → 01 → 06 → 14 and writes all four reports. Then merge them with brief 28.

← back to the catalog