gauntlet
Skill0 starsChanged 3 months agoMITView on GitHub
The anti-yes-man. Interrogate any idea, plan, price change, or pitch like an investor until it survives or dies, one pointed question at a time, each with a recommended answer. Run when the user says "is this a good idea", "poke holes in this", "run this through the gauntlet", "stress-test this", "would you invest in this", "play devil's advocate", or pitches something and asks what you think. Flattery is never the answer; this skill is.
20 other projects use a version of this.
Adopted by
Most recently taken up by
- laceyp99/skillstheir own version, first seen Nov 16, 2026
- tomw1808/flash-agentstheir own version, first seen Nov 15, 2026
- ryanportfolio/HarnessFirmware.comtheir own version, first seen Nov 8, 2026
- ryanportfolio/Corewise.Academytheir own version, first seen Nov 8, 2026
- ryanportfolio/savetokenstheir own version, first seen Nov 8, 2026
- ryanportfolio/STKtheir own version, first seen Nov 8, 2026
Used with
Projects with it also use addskill, apply-firmware and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
ryanportfolioSkilladdskill
Create, import, update, or install repo or personal skills: runtime ownership, resources, discovery checks.
Also in 6 projects that use this
ryanportfolioSkillapply-firmware
Bring current folder (plain folder, existing repo, old or partial firmware copy) up to full Harness Firmware in place. Mirror external repo → adopt-repo.
Also in 6 projects that use this
- +10 more from ryanportfolio
Other versions20 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 20
- oprogramadorreal/optimus-claude★ 75same name
- tomw1808/flash-agents★ 9same name
- bountyoperator/bounty-operator★ 6same name
- danecwalker/agentskills★ 2same name
- gioeleBrancaleoni/gauntlet★ 2same name
- Riloox/gauntlet-skill★ 2same name
- ryanportfolio/winuse-mcp★ 2same name
- bendyworks/claude-skills★ 1same name
- ryanportfolio/Corewise.Academy★ 1same name
- ryanportfolio/Harness-Firmware★ 1same name
10 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
We split this skill into the rules it gives an agent
7 rules. 0 are common, 4 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 18 edited versions we read keep these.
4
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 34
Read references/CONVENTIONS-core.md, then look for the charter (AGENTS.md / CLAUDE.md) and the vault.
- Distinctive: line 98
Before convening the board or writing a verdict, run the shared durable claim admission gate on every number, conversion, absence, and …
- Distinctive: line 109
Ask before you assert: never present a quantity, duration, sample size, date, or causal motive in prose before asking the founder for it or …
- Very specific: line 117
STOP. Read references/advisor-board.md now: it defines the five lenses, their bounded territories, and the merge rules.
- Distinctive: line 128
Every SURVIVES or DIES verdict is a consequential go/no-go recommendation, whatever the bet's size.
- Distinctive: line 165
After the verdict write, run the linked-state transaction: place the initiative link in the hub bucket matching its actual state, update …
- Very specific: line 171
Run scripts/graph-audit on every touched durable file in full mode.
“Common” is measured: kept by at least 6 in 10 of the 18 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
gauntlet: the idea earns its life here
A cheerleader with a knowledge cutoff is worthless to a founder. The gauntlet asks the questions an investor would, in the order a co-founder should, and it does not stop at the first comfortable answer. Weak ideas die in five minutes here instead of five weeks in the wild.
the pitch
|
[ mode check ] full collection present? read charter + vault first
|
[ size check ] quick poke ── small ideas, 2-4 questions
|
FULL GAUNTLET one question at a time, recommended answers,
| walk the idea's tree until nothing is left standing
| on assumption
[ deep tier ] BIG bets: convene the advisor board
|
VERDICT survives -> state: survived | dies -> buried with honorsStep 0: Mode check
Read references/CONVENTIONS-core.md, then look for the charter (AGENTS.md /
CLAUDE.md) and the vault. Apply the shared Universal preflight without losing the gauntlet's
explicit standalone exception.
- Full mode (the Ready signature is present): read the charter's opening business description, laws, and
glossary. Consult the vault for everything it knows about this idea's territory: past
attempts, related learnings, the vision file. Cite by file. The gauntlet attacks with THEIR
facts. Then locate or create the initiative file in
wiki/initiatives/BEFORE any interrogation: if new, it starts as a stub with frontmatter (type: initiative,title,state: idea,size,status: stated, dates),## The ideaholding the pitch, an empty## Gauntlet verdict, and## Log. Plan expands it to the full template later; the gauntlet owns only these headings. List a new stub inindex.mdandhub/initiatives.md. - Partial mode (some structure exists): name what is missing, stop and run the co-founder-setup skill before anything else in re-sync mode, and run the gauntlet verbally if the founder wants to continue. Write no files until setup repairs the signature.
- Standalone mode (nothing present): the gauntlet still runs, on interview-elicited facts
alone, verbally, writing no files. Ask the two grounding questions first (what is the
business, who pays). Quote the founder-confirmed answers in the transcript as the standalone
internal receipt, then gather the checked external leg before any go/no-go verdict. Label a
verdict
STANDALONE: SURVIVESorSTANDALONE: DIES: it expires with this conversation, writes no state, and cannot hand off to plan. Close by routing the founder to setup and a full re-run of the gauntlet before the bet enters the durable workflow. Missing external evidence uses the normalPARKED: external receipt missingoutcome.
Done when: you know the mode, hold whatever ground truth exists, and (full mode) the
initiative file exists at state: idea or better.
Step 1: Size the stakes
QUICK POKE for small, reversible, cheap ideas: 2-4 questions on the weakest points, verdict, done in minutes. FULL GAUNTLET for anything with real money, weeks of work, or customer contact. Uncertain resolves to the full gauntlet.
Done when: the tier is stated to the founder in one line.
Step 2: The interrogation
The rules, non-negotiable:
- One question per turn. A wall of questions produces diluted answers.
- Every question carries your recommended answer. The founder reacts to a position, never stares at a blank. You are proposing while interrogating; that is the co-founder posture.
- Facts are yours, decisions are theirs. Anything findable (in the vault, by research) you find; only judgment calls reach the founder.
- Attack with receipts. A counterclaim cites their vault or a checked external source. A doubt with no receipt gets asked as a question, never asserted as a fact.
- Walk the tree. Resolve dependent questions in dependency order: who pays, before what it costs; what it replaces, before how it grows.
The five holes to probe, in this order, skipping any the founder has already filled:
- Demand proof. What has a customer already PAID for or concretely done that says this is wanted? Interest is not evidence; behavior is.
- Workaround cost. What do those people do today instead, and what does it cost them in money or hours? No painful workaround, no urgency.
- Unit economics. What does one sale earn, what does it cost to get, and does the math survive at their real volume, not a fantasy volume?
- Smallest version. What is the least that still delivers the value? Everything beyond it is attachment, and attachment is where budgets die.
- Timing and durability. Why now, and does it still make sense if the obvious market shift happens? Push past answers every competitor could give.
A concrete founder gets few probes; a fuzzy one gets them all. The exit bar: every hole is either filled with substance or conceded as fatal. Do not soften a concession into a maybe.
Before convening the board or writing a verdict, run the shared durable claim admission gate on
every number, conversion, absence, and causal statement. Reopen the named source and quote the
supporting passage. Revenue cannot become runway without a founder-stated cash balance and burn
calculation; an empty SOP folder proves only that no SOP is recorded there. Founder memories stay
Founder-stated: unverified, including remembered attendee counts and buyer behavior.
Apply the shared parameter gate to every board assumption the vault does not supply: hourly cost,
delivery cap, customer alternative, buyer behavior, capacity, or scarcity. Ask it as the next
gauntlet question or label it Working hypothesis: unverified; never use it as a finding, required
modification, fatal hole, or verdict input.
Ask before you assert: never present a quantity, duration, sample size, date, or causal motive in prose before asking the founder for it or citing a receipt for it. Do not self-prime a fact (a runway, an attendee count) and then ask a leading question to have it confirmed.
Done when: nothing important is still standing on an assumption.
Step 3: The deep tier (BIG bets only)
STOP. Read references/advisor-board.md now: it defines the five lenses, their bounded
territories, and the merge rules. Improvised lenses overlap and dilute; the file is the panel.
Convene the board on the interrogation's survivors. Where the harness runs parallel agents, run the lenses in parallel; where it does not, run them as sequential passes with a fresh eye per lens. Merge per the reference.
Done when: each lens has reported inside its own territory and the merged findings are ranked.
Step 4: The verdict
Every SURVIVES or DIES verdict is a consequential go/no-go recommendation, whatever the bet's
size. Both receipt legs must exist: an internal receipt and a checked external source. A missing
leg replaces the go/no-go verdict with the shared canonical PARKED: internal receipt missing
or PARKED: external receipt missing, plus the recall or research action. Enthusiasm cannot
turn a parked recommendation into a go decision.
Verified absence can enforce a missing-prerequisite gate, but it cannot supply the external leg or
upgrade PARKED to DIES. State the operational refusal, preserve the idea state, and name the
evidence that would clear the gate.
In standalone mode, apply the shared exception explicitly: quote the transcript fact used as
the internal leg, name the external source, prefix the outcome STANDALONE:, and state that it
cannot advance or authorize work. Never present a standalone verdict as a durable gate result.
Three outcomes, delivered plainly. In full mode, EVERY outcome writes the complete durable gauntlet
record into the Gauntlet verdict section: outcome, full advisor board, strongest objections, founder
answers, exact receipt lines, and reopen condition. Then append
## [date] gauntlet | <title> to log.md; a died idea's reasoning is the
most valuable thing this skill produces, and it never lives only in chat.
Re-read the initiative state before the verdict write. An idea may advance to survived only
on SURVIVES. Re-running the gauntlet on survived work updates the latest verdict but keeps the
state; DIES or PARKED then blocks plan until a later SURVIVES. planned or executing work never
moves backward: record the pressure test and route a same-destination change to plan as an amendment;
a changed destination becomes a new idea. A reviewed or banked arc stays closed and
any renewed bet becomes a new idea with reopens: pointing to it.
- SURVIVES. Say what made it survive and what changed under fire. For an
idea, setstate: survived; every later state stays unchanged. Then write the verdict (verdict, strongest objection, what changed). Then the ONE next move: "say 'scope this out' and plan takes it from here." - DIES. Name the fatal hole in one sentence, honor what was good, and record what evidence would reopen it. State stays unchanged. Offer to bank a learning so the next idea starts smarter.
- PARKED. One missing fact decides it. Name the missing leg in the canonical shape, offer recall or research, state stays unchanged.
After the verdict write, run the linked-state transaction: place the initiative link in the hub bucket matching its actual state, update the index summary and current baton, and reconcile any queue entry that points at the verdict. A link that exists under the wrong bucket is stale.
Mandatory filing gate
- Run
scripts/graph-auditon every touched durable file in full mode. - Fix every violation and rerun it. A durable verdict, state change, or log row is incomplete until it exits zero. Standalone mode writes nothing and has no filing gate.
The founder should leave knowing exactly why, never just what.
Done when: the verdict is delivered with its reasons, and in full mode the initiative file, legal state, and log carry it whatever the outcome.
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.

