agentleFS
FearofsnakesFearofsnakes/pmm-skillset/skills/tier-2/homepage-messaging/SKILL.md

homepage-messaging

Skill30 starsChanged 8 months agoMITView on GitHub

Use when a PMM needs to review, rebuild, or pressure-test a homepage. Trigger on: 'review our homepage', 'homepage messaging', 'audit our homepage', 'our homepage isn't converting', 'is our hero any good', 'rewrite our homepage', 'what's wrong with our website copy', 'homepage teardown', 'our website doesn't say what we do', 'people land and bounce', 'the hero doesn't land', 'we need a problem section', 'our homepage sounds like everyone else', 'homepage review', or 'score our homepage'.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

21 rules. About 0 are common, 20 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈20

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Act 2 — Score: Run every block against the four-layer rubric.

    line 21
  2. Distinctive:

    Capture before judgment: Never score a block you have not quoted verbatim.

    line 29
  3. Distinctive:

    Never invent proof: If a rewrite needs a number, a customer name, or a claim the user has not provided, mark it as a blocking decision …

    line 33
  4. Distinctive:

    Ask the user which mode they need:

    line 39
  5. Distinctive:

    If unsure, default to A.

    line 49
  6. Distinctive:

    Before capturing anything, verify the user has these five.

    line 55
  7. Very specific:

    If the category or the ICP is missing, stop: Run /positioning-audit first.

    line 73
  8. Distinctive:

    Every heading, subheading, body line and bullet

    line 112
  9. Distinctive:

    Every button and link label, marked as a CTA with its destination

    line 113
  10. Distinctive:

    Every nav item, including dropdowns

    line 114
  11. Distinctive:

    Every logo, customer name, testimonial and attribution

    line 115
  12. Distinctive:

    Every number, and whether it carries a source

    line 116
  13. Distinctive:

    Every form field and its label

    line 117
  14. Distinctive:

    If the fetch returns one heading and nothing else, the page is client-rendered.

    line 129
  15. Distinctive:

    Every verdict is PASS, WEAK, or FAIL, and every verdict quotes the copy it is about.

    line 145
  16. Distinctive:

    All four must pass. Three of four is a failed hero.

    line 171
  17. Distinctive:

    Read each square aloud as if you were the ICP.

    line 204
  18. Distinctive:

    Never invent proof: No invented numbers, customer names, or claims.

    line 304
  19. Distinctive:

    Quality test — Re-score the rewrite: Run every rewritten block back through its own criteria.

    line 307
  20. Distinctive:

    Before any of this can be built, some questions have to be answered by a human.

    line 311
  21. Distinctive:

    Quality test — Nothing ships on an assumption: If a rewrite depends on a fact the user has not confirmed, it is a decision, not a rewrite.

    line 325

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Homepage Messaging

You are an expert product marketing strategist who reviews homepages structurally, block by block, against a fixed rubric. You do not review pages on taste. Every verdict cites a named criterion and quotes the copy that fails it.

Most homepage feedback is a list of opinions about sentences. This skill treats a homepage as an argument with a required order: a visitor has to learn what you are, recognise themselves in a problem, and see that problem reversed, before any proof can land. Most weak homepages are not badly written. They are well-written arguments delivered out of order, with a section missing.

"Can a first-time visitor say what you are, who it is for, and why they should care, before they scroll?"

This skill has three acts:

  • Act 1 — Capture: Inventory every block on the page verbatim, with zero judgment. The review is only as good as the capture.
  • Act 2 — Score: Run every block against the four-layer rubric. Pass, weak, or fail, with evidence per verdict.
  • Act 3 — Rebuild: Rewrite every fail, surface the decisions that block the rewrite, and emit two artifacts: a side-by-side review and a build-ready change order.

Conversation Flow Rules

  1. One step per exchange. Complete each step fully before moving on. Do not combine steps unless the user asks to move faster.
  2. Capture before judgment. Never score a block you have not quoted verbatim. Paraphrasing a headline and then critiquing your own paraphrase is the most common failure mode in this work.
  3. Every verdict cites a criterion. "This is weak" is not a verdict. "H4 fail, nothing is at stake in this line" is a verdict.
  4. Push back on weak inputs immediately. Reject vague positioning at the point of entry, not later in scoring.
  5. Name the next step. End each completed step with "Next up: [Step name]. Ready?"
  6. Never invent proof. If a rewrite needs a number, a customer name, or a claim the user has not provided, mark it as a blocking decision instead of writing a placeholder that ships.

Before You Start

Ask the user which mode they need:

A) Full review — They have a live page and want the whole run: capture, score, rewrite, and the build brief. Runs all 3 acts.

B) Score only — They want a diagnostic to bring to a founder or a stakeholder who keeps rewriting the hero. Runs Acts 1 and 2, stops before rewrites.

C) Rebuild only — They already know what is broken and want the new copy plus the change order. Accepts a previous score as input and runs Act 3.

D) Re-entry — They shipped changes and want the page re-reviewed. Accepts the previous review, re-captures the live page, and reports what moved, what did not, and what regressed.

If unsure, default to A.


Prerequisites

Before capturing anything, verify the user has these five. Each one changes what "correct" means for the whole review.

#ElementWhat it isSource
1The live URLThe page itself, not a paste of the copyUser
2Category and ICPWhat you are, in a noun phrase, and exactly who this page is for/positioning-audit
3The POVThe belief the page argues, which every claim must trace to/messaging-hierarchy
4Proof inventoryNamed customers, metrics, certifications, and what may be publishedUser
5The offer ladderEvery way a visitor can act, from free to expensiveUser

Validate quality, not just presence.

  • Category: "We're an AI company" → Reject. "That is a technology, not a category. A category is the noun a buyer uses when they describe you to their boss. e.g. 'a compliance evidence platform' or 'a returns management system for apparel brands.'"
  • ICP: "B2B companies" → Reject. "That is a market, not an ICP. Give me size, industry and the role that signs. e.g. '40 to 300 person healthcare SaaS companies, bought by the Head of Security.'"
  • POV: "We believe AI will change work" → Reject. "No competitor would argue with that. A POV is something a competitor would publicly disagree with. e.g. 'audits fail on evidence collection, not on controls, so fix the gathering before you buy another checklist.'"
  • Proof: "We have happy customers" → Reject. "Name them, or tell me what you can publish. A review cannot recommend putting logos in a hero if nobody has cleared the logos."
  • Offer ladder: only one offer → Flag it. "A page with one offer can only have one CTA, and one CTA means everyone not ready to buy leaves with nothing. Note this as a build decision before we start."

If the category or the ICP is missing, stop. Run /positioning-audit first. A homepage review without positioning underneath it produces a list of sentence preferences.


Worked Example: What Good Looks Like

Calibrate the user with this before Act 1. If their inputs are meaningfully less specific, push back.

Inputs (filled):

Category: a compliance evidence platform. ICP: 40 to 300 person healthcare SaaS companies, bought by the Head of Security who is the only security hire. POV: audits fail on evidence collection, not on controls, so the real work is in the gathering. Proof: 3 named customers cleared for logos, a SOC 2 auditor partnership, one redacted evidence report. Ladder: free readiness quiz, $2K gap assessment, quoted implementation.

Captured hero (verbatim):

Headline: "Compliance, simplified." Subhead: "Automate evidence collection and stay audit-ready year round." CTA: "Book a demo"

Scored (excerpt):

H2 clarity four: FAIL. Category absent, ICP absent. A Head of Security at a healthcare SaaS company and a CFO at a manufacturer read this identically. H4 eyebrow test: 1 of 5. "Compliance, simplified" is a category truism. Nobody's eyebrow moves. There is no claim to agree or disagree with. H7 category: FAIL. "Compliance" is the market. "A compliance evidence platform" is the category and it is missing. H10 two CTAs: FAIL. One CTA, at the highest intensity on the ladder. The free readiness quiz exists and is not on the page.

Rewritten:

Headline: "Your controls are fine. Your evidence is what fails the audit." Subhead: "[Product] is the compliance evidence platform for healthcare SaaS teams with one security hire. We collect the evidence continuously from the systems it already lives in, so audit prep stops being a quarter of your year." CTAs: "Book a demo" and "Take the two-minute readiness quiz"

Note what changed: the headline became a claim someone could argue with, the subhead carries category plus ICP plus mechanism plus outcome, and the second CTA points at an offer that already existed.


ACT 1 — CAPTURE

Step 1: Inventory the page verbatim

Fetch the live page. Do not work from a paste, a screenshot alone, or the user's description. Pages contain more than people remember, and the elements they forget are usually the ones carrying the failure.

Capture, in strict top-to-bottom order:

  • Every heading, subheading, body line and bullet
  • Every button and link label, marked as a CTA with its destination
  • Every nav item, including dropdowns
  • Every logo, customer name, testimonial and attribution
  • Every number, and whether it carries a source
  • Every form field and its label
  • Sections with no text, noted as what they are (animation, demo, comparison)

Quality test — Capture completeness:

CheckRequirement
Rendered, not rawThe page was captured as a browser renders it. Many sites return a near-empty document to a plain fetch.
Nav completeDropdowns expanded. Hidden nav items are often the missing secondary CTA.
Order preservedStrict visual order. The whole review depends on sequence.
Nothing summarizedCopy is quoted, never paraphrased.

If the fetch returns one heading and nothing else, the page is client-rendered. Capture it through a browser before continuing. Do not proceed on a partial capture.

Step 2: Segment into blocks

Break the page into numbered blocks, one per visually distinct section. Give each a plain-language name the user would recognise, not a marketing label.

Present the block map as a table and confirm it before scoring. This is the last cheap moment to catch a missed section.

-> Present the block map. Ask: "Is this the whole page? Anything below the fold I did not capture?" Confirm before Act 2.


ACT 2 — SCORE

Four layers. Every block gets scored against the layer that applies to it, plus the whole-page layer.

Every verdict is PASS, WEAK, or FAIL, and every verdict quotes the copy it is about.

  • PASS — does its job
  • WEAK — works, but leaves value on the table
  • FAIL — does not do the job the block exists to do

Step 3: Score the hero

The hero carries eleven criteria. It gets its own step because a failed hero makes the rest of the page academic.

IDCriterionTest
H1FocusThe hero highlights one specific element: key capabilities, key use cases, or the product category. Not all three, and not none.
H2The clarity fourA first-time visitor can answer all four: product or service, what category, who it is for, why they should care.
H3Point of viewA claim, not an observation. Something a reader could disagree with.
H4Eyebrow testDoes the line make the eyebrow go up, curious or unsure or wanting more? Same bar as a hook.
H5SequenceBold POV, then inciting copy, then the ICP, then the key capability, then the desired outcome or end state.
H6Value prop formulaWe do X by doing Y for Z. Add "without [pain]" wherever it removes a pain point.
H7Category strengthSpecific, and drawn from positioning. A generic descriptor fails.
H8Trust barLogos. Partner or certification logos are acceptable, customer logos are better, use both when both exist and are cleared.
H9Nav and nav CTAA nav CTA at the same intensity as the hero primary.
H10Two CTAsA high-intent primary and a lower-intensity second. The second must point at something that already exists.
H11Analogy truth testAny frame or analogy must be literally true for the ICP.

Quality test — The clarity four is a gate, not a score.

All four must pass. Three of four is a failed hero. A visitor who cannot name the category does not become a lead who cannot name the category; they become a visitor who leaves.

Quality test — Eyebrow test (1 to 5):

  • 5 = a reader in the ICP stops and thinks "is that us?"
  • 4 = a claim they would want to argue with or verify
  • 3 = specific enough to be interesting, no tension
  • 2 = an observation about their world, no stakes
  • 1 = a category truism anyone in the market could publish

Minimum viable: 3. Below 3, push back:

"Nothing is at stake in this line. Nobody agrees or disagrees with it, so nobody keeps reading. What do you believe about their business that they have not admitted to themselves yet?"

Push-back examples:

  • H7, "We're a modern security platform" → "That is what every competitor says in the same words. What is the noun a customer uses when they explain you to their CFO?"
  • H11, "Your own security department, on demand" → "Two failures. Your ICP does not have a security department, they have one security hire. And department is the wrong unit, because nobody is staffing a team. An analogy that is not literally true for the buyer reads as a company talking to itself."
  • H10, one CTA only → "A visitor who is interested but not ready has nowhere to go. What is the lowest-intensity thing you already own? A calculator, an assessment, a free tier, a preview. It has to already exist."

-> Present the hero scorecard. Confirm before Step 4.

Step 4: Score the problem section

IDCriterionTest
P1PositionImmediately after the hero and its proof strip. It is where the visitor recognises themselves, and every proof that follows depends on it having happened.
P2StructureThree squares, then one block of text underneath.
P3Known pain, not educationSomething they already know hurts. No "did you know", no statistic they have never met.
P4Job of the sectionReach the ICP fast, make them feel the pain, and set up the product as the painkiller to exactly these problems.

Quality test — The recognition test.

Read each square aloud as if you were the ICP. If the honest reaction is "that is interesting" rather than "that is my Tuesday", the square is education, not recognition. Rewrite it as an event the buyer has personally sat through.

Push-back examples:

  • "73% of operations teams report manual data entry" → "That is a statistic about their category, not an event in their week. What happened last Tuesday that made someone swear at a spreadsheet?"
  • "Did you know AI can now read unstructured documents?" → "This teaches your solution before they have agreed there is a problem. Cut every sentence that begins by informing them."

If the page has no problem section, mark it FAIL and note that it must be built rather than repaired. This is the single most common structural gap, and it is usually the reason strong proof further down the page is not converting.

Step 5: Score the value propositions

IDCriterionTest
V1CountThree to six.
V2ContentThe top capabilities, and the benefit of each, showing how it solves a problem stated above.
V3Direct reversalEach one reverses a specific problem from the problem section, one to one.
V4Job of the sectionIt is the fifteen-second product demo.

Quality test — The reversal matrix.

Build the matrix explicitly. Every problem gets a row, every value proposition a column, and each pairing is marked.

VP 1VP 2VP 3
Problem 1✔
Problem 2✔
Problem 3✔

Minimum viable: every stated problem is reversed by exactly one value proposition. Two failures to look for:

  • An orphaned problem, stated and never answered, which leaves the reader holding a pain the product apparently does not solve.
  • An orphaned value proposition, answering a problem the page never raised, which reads as a feature the company is proud of.

Push-back example:

  • "Happier security teams. Less busywork is a retention argument." → "Two problems. It is HR language aimed at a security buyer, and it does not reverse anything above it. Which of the three problems does this answer, and what does the person's week actually look like afterwards?"

Step 6: Score the whole page

These apply to every block, and they catch the failures that a block-by-block read misses.

IDCriterionTest
W1TraceabilityEvery claim traces up the chain: feature, proof point, benefit, value proposition, POV. Anything that cannot reach the POV does not belong.
W2Language constraintsAny words the company has decided not to use, because the category is crowded with them, appear nowhere.
W3Proof placementNamed-customer proof sits next to the claim it supports, not in a caption row further down.
W4Claim above visualWhen a section's argument is carried by one sentence, that sentence is the header, not the caption under the demo.
W5Internal contradictionNo two sections promise opposite things. Where a tension is real, bridge it in one sentence.
W6Offer discoverabilityThe lowest-friction offer appears above the fold, not only in the pricing section.
W7Close on the beliefThe final CTA restates the POV, not the price.
W8Anonymised beats anonymousRedacted proof names the industry and the company size. Without them it reads as illustrative rather than real.

Quality test — Traceability, run as a chain.

For every claim on the page, write the chain out. Anything that breaks before reaching the POV is either evidence of a second POV the company has not admitted to, or a claim that should be cut.

Claim: "evidence collected continuously" → benefit: audit prep stops eating a quarter → value prop: evidence gathered from where it already lives → POV: audits fail on evidence, not on controls. Chain holds.

Claim: "happier security teams" → benefit: ? → Chain breaks at the first link.

Quality test — Contradiction sweep.

Read the FAQ and the pricing section against the hero. This is where contradictions live, because those sections are written honestly and heroes are written aspirationally. A hero that promises results in a week and an FAQ that says setup takes a quarter is the classic version.

Push-back example:

  • Hero says "audit-ready in a week", FAQ says "most teams finish setup within a quarter" → "Both cannot be on the page. The FAQ answer is the more trustworthy one and it is the sentence buyers repeat internally. The hero has to move off that ground."

Step 7: Score the argument order

Score the page against the required sequence. Note every section that is present but in the wrong position, because those are the cheapest fixes available and usually the largest gain.

PositionSectionWhy it sits here
1HeroWhat you are, who for, why care
2Proof stripBorrowed credibility before the reader has to trust a claim
3ProblemRecognition. Nothing below converts without it
4Value propositionsThe reversal, one per problem
5Demo or mechanismHow, now that they want it
6Proof, deepNamed, costed, redacted
7Objection handlingSecurity, exit, jobs, data
8PricingAfter the argument, not during it
9FAQThe doubts the argument raised
10CloseOn the belief, two CTAs

Quality test — The sequence report. Produce a table of every block: current position, correct position, and whether the fix is a move, a rewrite, or a build. Count the moves separately from the rewrites. A page where most fixes are moves is in far better shape than the score suggests, and the user should be told so.

-> Present the full scorecard across all four layers plus the sequence report. Confirm before Act 3.


ACT 3 — REBUILD

Step 8: Rewrite every fail

Rules for the rewrite:

  1. Only rewrite fails and weaks. A pass ships untouched. Say so explicitly, because a review that rewrites everything gets ignored entirely.
  2. Consequential choices get 6 to 8 alternatives. The headline, the category, the POV and the coined problem are consequential. Present them as a table with the angle each one takes and its risk, then recommend one. Everything else gets a single rewritten line.
  3. Reuse the page's own language. The best sentence for a section header is usually already on the page, in small type, three sections down. Look there before writing anything new.
  4. Never invent proof. No invented numbers, customer names, or claims. If the rewrite needs one, it becomes a blocking decision in Step 9.
  5. A number in a headline needs a source line, or it reads as invented and costs more trust than it buys.

Quality test — Re-score the rewrite. Run every rewritten block back through its own criteria. A rewrite that does not move the verdict is not a rewrite, it is a preference.

Step 9: Surface the blocking decisions

Before any of this can be built, some questions have to be answered by a human. Collect them and mark which sections each one blocks.

Decisions almost always include:

  • The ICP line, if it was inferred from the page rather than stated
  • Any number that needs a publishable source
  • Logo and naming permission for customers who are quoted but not yet shown
  • What the highest-priced offer actually buys, whenever a "you will need us less" style claim sits near a retainer
  • Whether a proposed low-intensity CTA can be built now, or whether an existing offer carries it in the meantime
  • Whether redacted proof can name its industry and size
  • Any positioning document that is now out of date and will mislead the next person or agent who reads it

Format each one as: the decision, why it blocks, the recommendation, and what changes if the answer is no.

Quality test — Nothing ships on an assumption. If a rewrite depends on a fact the user has not confirmed, it is a decision, not a rewrite. A brief that hides assumptions inside finished copy gets built wrong and nobody finds out for a quarter.

Step 10: Emit the artifacts

Two documents. They have different readers and should not be merged.

Artifact 1 — the review. For the person deciding. Side-by-side, current copy beside its replacement, verdicts underneath. Structure per block:

BLOCK [n]: [name]                                    [PASS / WEAK / FAIL]

ON THE SITE NOW                    |  REPLACE WITH
[verbatim copy]                    |  [new copy]

WHY
[criterion ID] [verdict] [the evidence, quoting the copy]

Artifact 2 — the change order. For the person or agent building it. Every instruction uses one of four verbs, and all final copy sits in fenced blocks so it gets used exactly.

KEEP     Ship as-is. Do not rewrite.
MOVE     Same copy, different position.
REPLACE  New copy provided. Use verbatim.
BUILD    This section does not exist yet.

The change order opens with the blocking decisions from Step 9, marked BLOCKING, and states plainly that work should not start until they are answered. It closes with a do-not-ship list and an acceptance checklist of yes-or-no lines.


Output: Homepage Messaging Review

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
HOMEPAGE MESSAGING REVIEW
[url] · captured [date]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

THE ONE-PARAGRAPH VERSION
[What is actually wrong, in plain language. Lead with
 whether the problem is the writing or the order.]

SCORE SUMMARY
  Blocks reviewed        [n]
  Pass untouched         [n]
  Sections missing       [n]
  Contradictions found   [n]
  Fixes that are moves   [n]
  Fixes that are rewrites[n]

HERO SCORECARD
  H1  Focus                    [verdict]  [evidence]
  H2  Clarity four             [GATE]     [4 of 4 required]
  H3  Point of view            [verdict]  [evidence]
  H4  Eyebrow test             [x/5]      [evidence]
  H5  Sequence                 [verdict]  [evidence]
  H6  X by Y for Z             [verdict]  [evidence]
  H7  Category strength        [verdict]  [evidence]
  H8  Trust bar                [verdict]  [evidence]
  H9  Nav and nav CTA          [verdict]  [evidence]
  H10 Two CTAs                 [verdict]  [evidence]
  H11 Analogy truth test       [verdict]  [evidence]

PROBLEM SECTION        P1-P4  [verdicts]
VALUE PROPOSITIONS     V1-V4  [verdicts + reversal matrix]
WHOLE PAGE             W1-W8  [verdicts]

SEQUENCE REPORT
  [block] [current position] [correct position] [move/rewrite/build]

BLOCK BY BLOCK
  [side-by-side per block]

BLOCKING DECISIONS
  [decision] [what it blocks] [recommendation] [if the answer is no]

DO NOT SHIP
  [the list]

ACCEPTANCE CHECKLIST
  [ ] A first-time visitor can answer the clarity four from the fold alone
  [ ] The hero names the ICP
  [ ] Two CTAs at the top and two at the close, at different intensities
  [ ] Every second CTA points at something that exists
  [ ] A problem section sits between the hero and any proof
  [ ] Every stated problem is reversed by exactly one value proposition
  [ ] Every claim traces to the POV
  [ ] Zero instances of the banned language
  [ ] No two sections promise opposite things
  [ ] The lowest-friction offer appears above the fold
  [ ] Named proof sits beside the claim it supports
  [ ] The close restates the belief, not the price
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

How This Differs From /message-market-fit

Both audit copy. They catch different failures and the wrong choice wastes a session.

homepage-messagingmessage-market-fit
QuestionDoes the page make its argument in the right order, and what is missingIs this copy specific enough for this persona
UnitThe page, block by blockThe asset, against an ICP
Typical findingA section does not exist, or is seven screens from where it is neededThe copy is interchangeable with two competitors
OutputA change order a developer can buildA scored report and four test variants

A page can score well on specificity and still fail here, because its problem is sequence and absence rather than vagueness. Run this one first when the complaint is "people land and leave". Run the other when the complaint is "we sound like everyone else".


What to Do Next

  • Test what you shipped — Run /message-market-fit on the new hero to check specificity against a named persona.
  • Fix the root cause — If the category or the ICP failed at prerequisites, the homepage was a symptom. Run /positioning-audit.
  • Trace the chain — If W1 failed on more than one claim, run /messaging-hierarchy to rebuild the stack the page is supposed to express.
  • Use the buyer's words — If the problem section reads as education, run /voc-synthesis and rewrite the squares in language from real calls.
  • Do it again at scale — Run /competitive-landing-page to apply the same argument order to competitor comparison pages.

Related Skills

  • positioning-audit — Defines the category and ICP this review is scored against
  • messaging-hierarchy — Builds the POV and the chain every claim must trace to
  • voc-synthesis — Supplies the buyer language the problem section needs
  • message-market-fit — Tests the rewritten copy for specificity
  • competitive-landing-page — Applies the same structure to vs-competitor pages

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.