Every skill, tested
and documented.

You give itA draft email, memo, proposal, or status update, plus who it is for.

It returnsThe edited draft ready to send, a What-changed list, and flags for missing facts: bracketed placeholders, never guesses. Audit mode: names each pattern with the quoted line, no rewrite.

What verified meansRun by us from a clean install, checked against what it promises, documented here. Missing facts become bracketed placeholders, never guesses.

Changelogv1.0, Jul 2026. Verified release: edit and audit modes, corporate filler list, placeholders for missing facts.

You give itThe application, deck text, or email. Your rubric, if you have one.

It returnsA screening memo: snapshot, weighted score table, red flags, information gaps, and a verdict with its next step.

The mechanicsAnchored 1 to 5 scoring with evidence rules. Scores are never guessed, and integrity flags force an automatic pass.

Changelogv1.3, Jul 2026. This release: anchored rubric, evidence rules, automatic pass on integrity flags.

You give itThe screening memo, deal notes, or raw materials.

It returnsA materiality-ordered list: what would change the verdict, founder questions with priority tags, desk research and reference call tracks, and artifacts to request.

The mechanicsEvery question carries a source tag tracing it to the gap or flag it resolves.

Changelogv1.0, Jul 2026. First release: priority tags, source tags, coverage check.

You give itThe screening memo or question list, plus the call length.

It returnsA minute-budgeted 30 or 45 minute agenda: primary questions, pre-written follow-ups, flag probes, pre-call homework, and a post-call scorecard.

The mechanicsFollow-ups are pre-written against six named evasion patterns, like the deflection to vision.

Changelogv1.0, Jul 2026. First release: timed agenda, evasion-pattern follow-ups, post-call scorecard.

You give itThe completed screen, plus a sample or description of your house format.

It returnsA one-page memo in your sections and your verdict vocabulary, with scores, flags, and open items unaltered.

The mechanicsFormats and compresses only. It never re-scores and never softens a flag.

Changelogv1.0, Jul 2026. First release: house-format rendering, one-page cut order.

You give itThe deal file: screen, DD answers, call notes, data room extracts, financials, term sheets.

It returnsA house-format IC memo: indexed sources, every claim tagged, a bear-case risks section, and an open items list.

The mechanicsAssembles evidence, never manufactures conviction. Conflicting sources are shown side by side, never averaged.

Changelogv1.1, Jul 2026. This release: indexed source list, inline source tags, open items quarantine.

You give itThe memo with its pass reason, the founder's name, and the relationship context.

It returnsSend-ready email variants under 150 words each, plus an internal header with the real reason and the door-opener decision.

The mechanicsA shareability class on every reason. It never fabricates a reason and never adds false encouragement.

Changelogv1.2, Jun 2026. This release: warm, standard, and cold variants, reason shareability classes.

You give itThe draft, the audience, and one or two samples of your own writing if you want the voice matched.

It returnsThe rewrite ready to send, a What changed list naming each pattern it cut, and flags for anything that needs your judgment.

The mechanicsAn integrity check keeps every figure verbatim. Inconsistent numbers are flagged, never fixed, and bad news stays plain in both directions.

Changelogv1.0, Jul 2026. First release: seven-pattern taxonomy, figure integrity check, voice matching from samples.

You give itAn eight-field style brief, captured once. Then any write, rewrite, or check request.

It returnsThe finished copy, plus a rule-by-rule voice audit and flags on any claim the source does not support.

What verified meansRun by us from a clean install, checked against what it promises, documented here. Unsourced claims become bracketed placeholders, never invented facts.

Changelogv2.0, Jul 2026. Verified release: eight-field brief, visible audit on every output.

You give itOne essay, memo, newsletter, or talk transcript, plus the formats you want.

It returnsA claims inventory from the source, then per-format outputs each anchored to a claim, with a fidelity check.

What verified meansRun by us from a clean install, checked against what it promises, documented here. No new claims, and numbers are copied character for character.

Changelogv1.4, Jun 2026. Verified release: claims inventory, per-output fidelity check.

You give itThe transcript, recording notes, or fragmentary minutes.

It returnsA three-line summary, decisions with quoted evidence, an action-items table with owners and deadlines, open questions, and notable context.

What verified meansRun by us from a clean install, checked against what it promises, documented here. Missing owners and deadlines are marked, never guessed.

Changelogv3.1, Jul 2026. Verified release: decision test, quoted evidence, action-items table.

You give itThe founder update emails or notes, plus last period's report if you have it.

It returnsPer-company blocks in a fixed schema: cash, runway, core metric with trend, asks, risks, and status, plus a portfolio roll-up with cross-company themes.

The mechanicsStatus rules read the numbers, not the founder's mood. Nothing is carried forward or inferred.

Changelogv1.0, May 2026. First release: fixed company schema, portfolio roll-up.

You give itYour thesis and mandate, set once. Then any deal materials.

It returnsA meets, misses, or unclear scorecard across eight mandate criteria, an in, edge, or out verdict, one batched request list, and a next step.

The mechanicsNever scores against an imagined mandate. Ownership and fund-returns arithmetic run as one-line checks.

Changelogv1.0, Jun 2026. First release: eight-criterion scorecard, exception clause for edge cases.

You give itThe job posting and your background, plus anything you know about the company and the interview stage.

It returnsA one-page prep sheet: two to three stories in the Stakes, Move, Result, Speed, Proof format, answers mapped to their problem, gap reframes, a research hook, and a closing question.

The mechanicsEvery result carries a timeframe, every claim is ownership your input supports, and gap reframes change the story, never the facts.

CreditAdapted from Jordan Carver's Startup Interview Cheat Sheet, a one-page guide from the interviewer's side of the table.

Changelogv1.0, Aug 2026. First release: story bank, gap reframes, do-not-lead-with checks, closing question.

You give itA market, category, or company question, plus any materials you already have.

It returnsA decision-oriented research brief: sizing, competitive landscape, and a summary with source attribution throughout.

What verified meansRun by us from a clean install, checked against what it promises, documented here. Figures arrive with their sources; anything unsourced is flagged, not asserted.

CreditBuilt by Affaan Mustafa, packaged from his ECC repo with license and attribution included (MIT).

Changelogv1.0, Aug 2026. First verified release, packaged unmodified from the source repo.

You give itA company name plus whatever you have: deck, notes, public links.

It returnsA readout per perspective and a cross-perspective synthesis, so you see where the three seats disagree.

What verified meansRun by us from a clean install, checked against what it promises, documented here. Gaps in the record are surfaced as open questions, not filled in.

CreditBuilt by Alex Yang, packaged from his finance-skills repo with license and attribution included (MIT).

Changelogv1.0, Aug 2026. First verified release, packaged unmodified from the source repo.

You give itThe company or thesis everyone at the table already likes, plus the materials behind it.

It returnsThe case against: counterevidence, failure modes, and the questions a skeptic would put to the sponsor.

What verified meansRun by us from a clean install, checked against what it promises, documented here. It argues one side only by design; the bull case stays your job.

CreditBuilt by YenNan Liu as the bear-case skill in his InvestSkill repo, packaged here under the name we list it as (MIT).

Changelogv1.0, Aug 2026. First verified release, packaged from the source repo with the skill content unmodified.

You give itA product or category, then a short intake interview to anchor the work.

It returnsModule-based briefs in a repeatable template: competitors, avatar, market, and value proposition.

What verified meansRun by us from a clean install, checked against what it promises, documented here. Outputs are framed as research to pressure-test, not facts to paste.

CreditBuilt by Ishwar Jha, packaged with license and attribution included (Apache 2.0).

Changelogv1.0, Aug 2026. First verified release, packaged unmodified from the source repo.

You give itA product idea at any stage before build, and honest answers as it pushes back.

It returnsA worked session across the six questions, from demand reality to the narrowest wedge, saved as a design doc.

What verified meansRun by us from a clean install, checked against what it promises, documented here. Built for Claude Code; a few extras assume the author's own setup and sit idle elsewhere.

CreditBuilt by Garry Tan as part of his GStack setup, packaged with license and attribution included (MIT).

Changelogv1.0, Aug 2026. First verified release, packaged unmodified from the source repo.

You give itA plan or roadmap that feels safe, and a mode: expand scope, cherry-pick, hold, or strip back.

It returnsA challenge to the plan's premises with an explicit scope decision and the reasoning behind it.

What verified meansRun by us from a clean install, checked against what it promises, documented here. Built for Claude Code; a few extras assume the author's own setup and sit idle elsewhere.

CreditBuilt by Garry Tan as the plan-ceo-review skill in his GStack setup, packaged here under the name we list it as (MIT).

Changelogv1.0, Aug 2026. First verified release, packaged from the source repo with the skill content unmodified.

You give itA hypothesis, the variants to test, and a metric; then data points as content ships.

It returnsStatistical winners, a living playbook of proven rules, and suggestions for the next experiment.

What verified meansRun by us from a clean install, checked against what it promises, documented here. The pacing alerts assume your own platform API credentials; the experiment engine runs without them.

CreditBuilt by Eric Siu at Single Grain, packaged with license and attribution included (MIT).

Changelogv1.0, Aug 2026. First verified release, packaged from the source repo with install frontmatter added and the repo-level telemetry preamble removed; the methodology and scripts are unmodified.