Index of every page on this site: llms.txt
All skills
aidevgtm avatar

/17-review-the-work

@0e9e3d4
Run it yourself

Before showing the user any substantive GTM deliverable (positioning, value prop, homepage, launch post, pricing, sales script, the brief, the market research, or the roadmap), stress-test it against the standard as an independent critic, because the agent that wrote it is the worst judge of whether it is good. Use as a gate right before presenting work, or when the user asks whether something is actually strong.

  • 1 file
  • 5.6 KB
  • Updated last month
  • GitHub
Use this Skill: https://skilld.dev/gh/aidevgtm/gtm-cofounder/17-review-the-work

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈109 tokens for metadata: the name and description. ≈1.3k when used: this file.

Description uses 5.4% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈434 characters, or 5.4%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Review the work (the self-check gate)

The person who wrote it is the worst judge of whether it is good. Before this reaches the founder, stop being the author and become the skeptic.

Use this when: you are about to present any substantive deliverable, or the founder asks "is this actually good?" It is a gate, not a stage. It has no place in the linear sequence. Run it around any piece of work, every time.

The core idea

An agent that just wrote something is biased to ship it. That bias is how generic, plausible, quietly wrong work reaches a founder who does not yet know enough to catch it. So separate the two jobs: the author drafts, a different lens judges. You do not present work because you made it. You present it because it survived a skeptic.

Steal the one rule that makes this work: the author may submit the work, the author may not issue the verdict. Switch roles on purpose. Become the developer who is skeptical, the buyer who is busy, the reviewer who has seen a hundred of these, and try to break it before the market does.

How to run this (for the agent)

  • Run it silently, before presenting. The founder should see the verdict and the work, not the whole audit.
  • Give a clear verdict: PASS, or REVISE with the exact checks that failed and the specific fix for each. Never a vague "looks good."
  • Do not rubber-stamp. If you cannot find a single weakness, you are not reading as the skeptic. Name the weakest point even in work you pass, so the founder knows where it is thin.
  • When it fails, fix it and re-run the gate, then present. Do not hand the founder a list of problems you could have fixed yourself.

The standard (applies to everything)

  • Specific, not puffery. No "powerful," "seamless," "best-in-class," "platform." Every claim carries a number, a name, or a proof, or it gets cut.
  • The developer is the hero, not the product. If the founder's tool is the hero of the sentence, rewrite it.
  • It speaks to the real ICP from the brief, not to "developers." A message for everyone lands on no one.
  • It rests on validated facts. Anything load-bearing that is still [assumption] is flagged out loud, not smuggled in as truth.
  • Every outside fact is real and linked. A community, a member count, a launch result, a quote: each one has a link you actually checked. If you can't point to the source, cut it or mark it unknown. An invented fact the founder acts on is the worst failure in this pack.
  • A skeptical developer could not immediately prove it false. Developers test claims and find the truth. If a dev would find the failing case in thirty seconds, you have not passed.
  • Human voice. Reads like a person wrote it. No em-dashes, no AI tells.

The specific checks (by deliverable)

Run the general standard, then the relevant skill's own checklist:

  • Positioning / story (positioning-and-story): problem-first, not solution-first. Has a villain. Survives the "every competitor could say this" test.
  • Value prop (value-prop-that-converts): no banned words, built on a real job-to-be-done, and a developer would repeat it in their own words.
  • Homepage (the-homepage): written for the developer who visits, not the buyer who never does. Time to first value is obvious.
  • Launch post (launch-it): honest, not salesy. Opens on the developer's problem, not the product. Would not get flamed for overclaiming.
  • Pricing (pricing): anchored to a value metric, not a guessed number. The buyer could justify it to whoever holds the budget.
  • Founder-led sales (founder-led-sales): passes the four-part deal test. Does not mistake a friendly free user for a buyer.
  • The brief (start-here): strongest asset at the top, [validated] / [assumption] tags intact, in the founder's own words.
  • The market research (research-the-market): leads with what the founder didn't say. Venues are named and linked, ranked by fit not size, with self-promo rules noted. Nothing is filled in from memory.
  • The roadmap (strategy-and-roadmap): opens with a one-line Diagnosis, then What the research added, then Now / Next / Later. It is a plan, not a re-saved copy of the brief. The founder-facing test: would they read it and learn something they didn't tell you? If every move traces back to their own answers, it fails. Channel and launch moves name specific venues, not categories.

Your next 30 minutes

  • Take the last thing produced (a headline, a post, a price) and run it through the standard above, as the skeptic, not the author.
  • Name the single weakest claim in it. Replace it with something provable, or cut it.
  • Find the one [assumption] it depends on most, and decide whether it is safe to ship on or belongs in talk-to-users first.
  • Only then put it in front of a real person. If it did not survive you, it will not survive them.

Built from real dev-tool GTM experience, with frameworks from Adam Frankl (The Developer-Facing Startup) and Jakub Czakon (markepear.dev). Want an independent human read before this goes live? That's exactly what I do. Reach out to me on LinkedIn.

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No third-party reports yet.

Signed by skilld at 0e9e3d4. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 4 days ago.

Activeupdated last month

README badge

README badge for aidevgtm/gtm-cofounder/17-review-the-work