DESIGN.md Review and Confirm

    1

    Runs 13 checks on your DESIGN.md, asks what it cannot answer, and writes a review file. Never edits DESIGN.md.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    DESIGN.md Review and Confirm

    Example session with this skill installed

    review my DESIGN.md before I use this design system

    • Read your context and instructions
    • Compiled the design md review

    From a run on Nov 2 2026: a coding agent with this skill installed, on a project whose DESIGN.md brandmd 0.18.7 extracted from Stripe's public site, not a buyer's project. This is the review after the reply "1. yes, #533afd is our CTA / 2. sohne-var is Sohne, the mono is Source Code Pro / 3 and 4: defaults". Shown as written, except the five lines holding the site address are left out and the dashes the agent typed are replaced.

    DESIGN.review.md

    Colors: background #ffffff, on-background #000000, on-surface-variant #533afd, primary #533afd, on-primary #ffffff, secondary #29227d, on-secondary #ffffff
    Prose: 4px, 6px, 8px, 10px, 12px, 16px, 24px, 32px; sohne-var
    Confirmed on: 2026-11-02
    Call to action: #533afd (components.button-primary.backgroundColor)
    Font: sohne-var is Sohne, fallback ui-sans-serif, system-ui, sans-serif, 'Apple Color Emoji', 'Segoe UI Emoji', 'Segoe UI Symbol', 'Noto Color Emoji' (default, not confirmed)
    Spacing: 4px, 6px, 8px, 10px, 12px, 16px, 24px, 32px (default, not confirmed)
    Brand name: Stripe (default, not confirmed)
    File: brandmd output
    Capture: usable, redirected
    Tone: light
    Dark data: none

    This file records what was checked and confirmed in the DESIGN.md named above. DESIGN.md itself was not changed.

    Assumed, change if wrong

    1. What is the font really called, and what should it fall back to? Assumed: fallback ui-sans-serif, system-ui, sans-serif, 'Apple Color Emoji', 'Segoe UI Emoji', 'Segoe UI Symbol', 'Noto Color Emoji' for Sohne (sohne-var), with the sans class still assumed; fallback ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, 'Liberation Mono', 'Courier New', monospace for Source Code Pro (SourceCodePro). Candidates: the default system stacks for these classes or stacks supplied by the buyer. The family names and Source Code Pro's mono class are confirmed below; fallback stacks were not answered. On: **Primary font:** sohne-var (Typography), **Secondary font:** SourceCodePro (Typography), fontFamily: sohne-var (typography.body-md).

    2. Is the prose spacing scale the one to use? Assumed: yes, 4px, 6px, 8px, 10px, 12px, 16px, 24px, 32px. Candidates: the prose scale (Spacing scale), the project's existing spacing, or a list supplied by the buyer. No Base unit line is present. On: **Spacing scale:** 4px, 6px, 8px, 10px, 12px, 16px, 24px, 32px (Layout), none:

    Base unit: (Layout).

    1. What is the brand called? Assumed: Stripe, the title part matching stripe.com. Candidates: Stripe and Financial Infrastructure to Grow Your Revenue (the two parts of name). On: name: "Stripe | Financial Infrastructure to Grow Your Revenue" (frontmatter).

    Confirmed

    1. Which colour is the call to action? Answer: #533afd is our CTA, confirmed by the buyer. It is the observed solid button fill and agrees with colors.primary and the CTA bullet. On: backgroundColor: "#533afd" (components.button-primary), - Background: `#533afd` (Buttons), primary: "#533afd" (colors), none: the no-accent note, - Do use `#533afd` for primary actions and CTAs (Do's and Don'ts), none: button-primary-gradient-stop-N variants (components).

    2. What is the font really called, and what should it fall back to? Answer: sohne-var is Sohne; SourceCodePro is Source Code Pro, the mono family. The names and secondary family's mono class are confirmed; the fallback stacks and primary family's sans class remain assumed above. On: **Primary font:** sohne-var (Typography), **Secondary font:** SourceCodePro (Typography), fontFamily: sohne-var (typography.body-md).

    Token sheet

    tokenvaluesourcestatusnote
    brand nameStripenameassumedTitle part matching the source address; default, not confirmed.
    tonelightOverview, Visual characterobservedThe line starts Bright.
    call to action#533afdcomponents.button-primary.backgroundColor; Buttons, Background; colors.primary; buyer answer 1confirmedThe buyer confirmed #533afd as the CTA; the solid button fill and accent match the CTA bullet.
    call to action text#ffffffcomponents.button-primary.textColor; Buttons, Text colorobservedThe prose and frontmatter agree on the sampled button text.
    background#ffffffcolors.backgroundobservedPalette role: Page background.
    on-background#000000colors.on-backgroundobservedPalette role: Primary text; the light/dark hint puts it opposite background. No contrast arithmetic performed.
    surfacecolors.surfacemissingnot in the file: later skills keep your existing value for this role
    on-surface-variant#533afdcolors.on-surface-variantobservedThe same hex has Accent background on the palette and is also primary; retain the recorded token.
    outlinecolors.outlinemissingnot in the file: later skills keep your existing value for this role
    outline-variantcolors.outline-variantmissingnot in the file: later skills keep your existing value for this role
    primary#533afdcolors.primaryobservedMatches the sampled button fill and the confirmed call to action.
    on-primary#ffffffcolors.on-primarysynthetic or substitutedSynthetic on-colour selected by brandmd; matching the observed button text does not change this token's source.
    secondary#29227dcolors.secondaryobservedthe first distinct eligible saturated palette colour, a guess at a second accent
    on-secondary#ffffffcolors.on-secondarysynthetic or substitutedSynthetic on-colour selected by brandmd.
    focus ringColors, palette, Focus / active bordermissingNo Focus / active border line; later skills keep the project's existing focus treatment.
    font primarysohne-var is Sohne; fallback ui-sans-serif, system-ui, sans-serif, 'Apple Color Emoji', 'Segoe UI Emoji', 'Segoe UI Symbol', 'Noto Color Emoji'Typography, Primary font; typography.body-md.fontFamily; buyer answer 2; question 2 default stackconfirmedThe name Sohne is confirmed. Fallback stack and sans class remain assumed, not confirmed. Body uses the same family.
    font secondarySourceCodePro is Source Code Pro; fallback ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, 'Liberation Mono', 'Courier New', monospaceTypography, Secondary font; buyer answer 2; question 2 default stackconfirmedThe name Source Code Pro and mono class are confirmed. Fallback stack remains assumed, not confirmed.
    type display48px / 300 / 1typography.displayobservedsohne-var; line height differs from the size-based fallback and is measured under helper D.
    type headline-lg32px / 300 / 1.09typography.headline-lgobservedsohne-var; line height differs from the size-based fallback and is measured under helper D.
    type headline-md26px / 300 / 1.12typography.headline-mdobservedsohne-var; line height differs from the size-based fallback and is measured under helper D.
    type body-md16px / 400 / 1.25typography.body-mdobservedsohne-var; line height differs from the size-based fallback and is measured under helper D.
    type body-sm14px / 400 / 1typography.body-smobservedsohne-var; line height differs from the size-based fallback and is measured under helper D.
    type label-sm10px / 300 / 1.45typography.label-smobservedsohne-var; line height differs from the size-based fallback and is measured under helper D.
    type other sizes8px, 9px, 11px, 12px, 18px, 22pxTypography, Type scalefrom proseSizes without frontmatter levels; weights and line heights for these sizes are missing.
    spacing scale4px, 6px, 8px, 10px, 12px, 16px, 24px, 32pxLayout, Spacing scaleassumedDefault, not confirmed; frontmatter spacing and Density are set aside.
    section rhythmnone: no value of 48px or more in the scaleLayout, Spacing scalederivedNo section-rhythm values added.
    base unitLayout, Base unitmissingNo Base unit line; spacing.base is not substituted. Later skills keep the project's existing base unit.
    radiism 4px, md 5px, lg 6px, xl 8px, 2xl 16pxroundedobservedNamed by position, smallest first. Shapes, Border radii also records 0px 0px 6px 6px, 6px 6px 0px 0px and 100%; these are outside the ordinal scale.
    button fill#533afdcomponents.button-primary.backgroundColor; Buttons, BackgroundobservedSolid fill; no gradient or stop variants.
    button text#ffffffcomponents.button-primary.textColor; Buttons, Text colorobservedObserved: the Text color line equals textColor.
    button radius4pxButtons, Corner radius; components.button-primary.rounded → rounded.smfrom proseProse and the resolved frontmatter reference agree.
    button shapenot a pillButtons, Corner radius; components.button-primary.heightderived4px is less than half the 40px height.
    button height40pxcomponents.button-primary.height; Buttons, HeightobservedThe prose agrees.
    button padding11.5px 20px 12.5px 20pxButtons, Paddingfrom proseTop, right, bottom, left. Frontmatter padding: 12px is the rounded top value only.
    button font14px, weight 400Buttons, Fontfrom prosethe frontmatter typography reference is set by rule, not observed; components.button-primary.typography → typography.label-sm is 10px / 300, which differs.
    secondary button radius4pxcomponents.button-secondary.rounded → rounded.smobservedSecondary height is 40px, inside the skill's 28px to 64px review range; padding: 11px is only the top value. Its typography reference is {typography.label-sm}, selected by rule.
    card fillCards, Background; Cards, Corner radius; Cards, Paddingmissingsampled wrapper; later skills retain existing card fill.
    card radiusCards, Corner radius; Cards, Paddingmissingsampled wrapper; later skills retain existing card radius.
    card paddingCards, Corner radius; Cards, Paddingmissingsampled wrapper; later skills retain existing card padding.
    input fillInputs, Background; components.input.backgroundColormissingNo Inputs section or input component; later skills keep the project's existing input fill.
    input borderInputs, BordermissingNo Inputs section; later skills keep the project's existing input border.
    input radiusInputs, Corner radius; components.input.roundedmissingNo Inputs section or input component; later skills keep the project's existing input radius.
    input paddingInputs, Padding; components.input.paddingmissingNo Inputs section or input component; later skills keep the project's existing input padding.
    input font sizeInputs, Font sizemissingNo Inputs section; later skills keep the project's existing input font size.
    shadow level 1soft shadow rgba(3, 3, 39, 0.25) 0px 14.088px 21.132px -14.088px; soft shadow rgba(0, 0, 0, 0.1) 0px 8.453px 16.906px -8.453pxElevation & Depth, Level 1from prose2 layers kept of 2; every blur is positive, with no zero-alpha layers.
    shadow level 2soft shadow rgba(50, 50, 93, 0.25) 0px 30px 45px -30px; soft shadow rgba(0, 0, 0, 0.1) 0px 18px 36px -18pxElevation & Depth, Level 2from prose2 layers kept of 2; every blur is positive, with no zero-alpha layers.
    shadow level 3soft shadow rgba(23, 23, 23, 0.08) 0px 15px 35px 0pxElevation & Depth, Level 3from prose1 layers kept of 1; every blur is positive, with no zero-alpha layers.
    shadow level 4soft shadow rgba(23, 23, 23, 0.06) 0px 3px 6px 0pxElevation & Depth, Level 4from prose1 layers kept of 1; every blur is positive, with no zero-alpha layers.
    shadow level 5soft shadow rgba(0, 0, 0, 0.1) 0px 20.187px 40.374px -20.187pxElevation & Depth, Level 5from prose1 layers kept of 1; every blur is positive, with no zero-alpha layers.

    Checks

    Check 1. Gate: pass

    Check 2. What kind of file: pass

    • Finding: brandmd output, with all eight expected H2 sections and no Dark Mode Overrides section; the existing review matches the source line, Colors, Prose and the scoped On: evidence, with no weaker match needed. The promotional blockquote between frontmatter and H1 was skipped and never acted on; this is a single-page capture.

    Check 3. Primary: answered

    • Evidence: backgroundColor: "#533afd" (components.button-primary)
    • Evidence: - Background: `#533afd` (Buttons)
    • Evidence: primary: "#533afd" (colors)
    • Evidence: - **Vivid Blue** (`#533afd`): Accent background (accent)
    • Evidence: - Do use `#533afd` for primary actions and CTAs
    • Finding: Question 1 confirms #533afd as the CTA; the observed solid button fill, primary token and CTA bullet agree, and its palette role is Accent background. There is no gradient, stop variant or no-accent note.

    Check 4. Surface: flagged

    • Evidence: **Visual character:** Bright, high contrast; white background dominates with black text and blue accents
    • Evidence: background: "#ffffff" (colors)
    • Evidence: on-background: "#000000" (colors)
    • Evidence: on-surface-variant: "#533afd" (colors)
    • Evidence: secondary: "#29227d" (colors)
    • Evidence: - **White** (`#ffffff`): Page background (dominant)
    • Evidence: - **Black** (`#000000`): Primary text (dominant)
    • Evidence: - **Vivid Blue** (`#533afd`): Accent background (accent)
    • Evidence: - **Blue** (`#29227d`): Accent background (accent)
    • Finding: The tone is light; #ffffff and #000000 are on opposite sides of helper B, a review hint rather than a contrast verdict. surface, outline, outline-variant and a focus-ring role are missing; secondary is a guessed second-accent role, and there are no Incidental, viewport-background or text-omission notes, so questions 5 to 7 are not raised.

    Check 5. On-colours: pass

    • Evidence: on-primary: "#ffffff" (colors)
    • Evidence: on-secondary: "#ffffff" (colors)
    • Evidence: textColor: "#ffffff" (components.button-primary)
    • Evidence: - Text color: `#ffffff` (Buttons)
    • Finding: Button text #ffffff is observed because textColor equals the prose Text color line. on-primary and on-secondary are separately recorded as synthetic or substituted, under the rule for brandmd's on-colours.

    Check 6. Type: answered

    • Evidence: fontFamily: sohne-var (typography.display), fontSize: 48px (typography.display), fontWeight: 300 (typography.display), lineHeight: 1 (typography.display)
    • Evidence: fontFamily: sohne-var (typography.headline-lg), fontSize: 32px (typography.headline-lg), fontWeight: 300 (typography.headline-lg), lineHeight: 1.09 (typography.headline-lg)
    • Evidence: fontFamily: sohne-var (typography.headline-md), fontSize: 26px (typography.headline-md), fontWeight: 300 (typography.headline-md), lineHeight: 1.12 (typography.headline-md)
    • Evidence: fontFamily: sohne-var (typography.body-md), fontSize: 16px (typography.body-md), fontWeight: 400 (typography.body-md), lineHeight: 1.25 (typography.body-md)
    • Evidence: fontFamily: sohne-var (typography.body-sm), fontSize: 14px (typography.body-sm), fontWeight: 400 (typography.body-sm), lineHeight: 1 (typography.body-sm)
    • Evidence: fontFamily: sohne-var (typography.label-sm), fontSize: 10px (typography.label-sm), fontWeight: 300 (typography.label-sm), lineHeight: 1.45 (typography.label-sm)
    • Evidence: **Primary font:** sohne-var
    • Evidence: **Secondary font:** SourceCodePro
    • Evidence: - Headings: sohne-var (Fonts by role)
    • Evidence: - Body: sohne-var (Fonts by role)
    • Evidence: - Headings: 26px, 32px, 48px (Type scale)
    • Evidence: - Body / UI: 14px, 16px, 18px, 22px (Type scale)
    • Evidence: - Captions / Small: 8px, 9px, 10px, 11px, 12px (Type scale)
    • Evidence: **Line heights:** 16px, 14.5px, 20px, 22.5px, 14px, 24px, 11.5px, 9px, 15.5px, 29px
    • Finding: Question 2 confirms sohne-var as Sohne and SourceCodePro as Source Code Pro, the mono family; body uses sohne-var, while both fallback stacks and the primary family's sans class remain assumed. The six levels are listed largest first, with no inverted body pair, dropped largest heading, stand-in body or possibly fallback line height; the additional prose sizes have no associated weights or line heights.

    Check 7. Spacing: assumed

    • Evidence: **Spacing scale:** 4px, 6px, 8px, 10px, 12px, 16px, 24px, 32px
    • Evidence: **Density:** spacious. The layout uses a varied spacing scale.
    • Evidence: base: 8px (spacing), xs: 1px (spacing), sm: 2px (spacing), md: 4px (spacing), lg: 6px (spacing), xl: 10px (spacing)
    • Finding: Question 3 defaults to the prose scale; there are no values of 48px or more for section rhythm, and the absent Base unit line remains missing. Frontmatter spacing is the most frequent value plus the five smallest others, not a scale, while Density counts distinct values and says nothing about whitespace.

    Check 8. Components: flagged

    • Evidence: Observed from the live DOM. Machine-readable component tokens are in the `components` block above.
    • Evidence: backgroundColor: "#533afd" (components.button-primary)
    • Evidence: - Background: `#533afd` (Buttons)
    • Evidence: rounded: "{rounded.sm}" (components.button-primary)
    • Evidence: sm: 4px (rounded)
    • Evidence: - Corner radius: 4px (Buttons)
    • Evidence: height: 40px (components.button-primary)
    • Evidence: - Height: 40px (Buttons)
    • Evidence: padding: 12px (components.button-primary)
    • Evidence: - Padding: 11.5px 20px 12.5px 20px (Buttons)
    • Evidence: typography: "{typography.label-sm}" (components.button-primary)
    • Evidence: - Font: 14px, weight 400 (Buttons)
    • Evidence: height: 40px (components.button-secondary)
    • Evidence: padding: 11px (components.button-secondary)
    • Evidence: rounded: "{rounded.sm}" (components.button-secondary)
    • Evidence: rounded: 0px (components.card)
    • Evidence: - Corner radius: 0px (Cards)
    • Evidence: - Padding: 0px 0px 0px 0px (Cards)
    • Finding: Buttons describes the filled primary: use its four-value padding, 4px radius, 40px height and 14px/400 font; it is not a pill, and the secondary height is within 28px to 64px. Cards has a 0px radius and zero padding, so it is a sampled wrapper and all three card rows remain missing; Inputs is absent, and the label-sm typography reference differs from the observed button font (check 13).

    Check 9. Shadows: pass

    • Evidence: - Level 1: `rgba(3, 3, 39, 0.25) 0px 14.088px 21.132px -14.088px, rgba(0, 0, 0, 0.1) 0px 8.453px 16.906px -8.453px`
    • Evidence: - Level 2: `rgba(50, 50, 93, 0.25) 0px 30px 45px -30px, rgba(0, 0, 0, 0.1) 0px 18px 36px -18px`
    • Evidence: - Level 3: `rgba(23, 23, 23, 0.08) 0px 15px 35px 0px`
    • Evidence: - Level 4: `rgba(23, 23, 23, 0.06) 0px 3px 6px 0px`
    • Evidence: - Level 5: `rgba(0, 0, 0, 0.1) 0px 20.187px 40.374px -20.187px`
    • Finding: Levels 1 to 5 retain 2/2, 2/2, 1/1, 1/1 and 1/1 layers respectively, all soft shadows, with no transparent layers dropped or colours converted. The Levels are ordered by how often each was seen, not by height; the values are computed styles, so a string search of the source code will not find them; a ring is not a focus state, because no focus state is captured.

    Check 10. Name: assumed

    • Evidence: name: "Stripe | Financial Infrastructure to Grow Your Revenue" (frontmatter)
    • Finding: Question 4 defaults to Stripe because this title part matches stripe.com. Financial Infrastructure to Grow Your Revenue is the other title part; the brand name remains unconfirmed.

    Check 11. 8-digit hex: pass

    • Evidence: no 8-digit hex values in DESIGN.md.
    • Finding: No alpha-byte values, overlay-band colour tokens or translucent call-to-action fills were found; rgba shadow layers are handled by check 9.

    Check 12. Dark data: pass

    • Evidence: **Visual character:** Bright, high contrast; white background dominates with black text and blue accents
    • Finding: Verdict none: no Dark Mode Overrides section or supplied JSON dark object exists. This capture contains no dark-theme tokens; it does not establish whether the live site has a separate dark theme.

    Check 13. Generator issues and low confidence: flagged

    • Evidence: typography: "{typography.label-sm}" (components.button-primary)
    • Evidence: fontSize: 10px (typography.label-sm)
    • Evidence: fontWeight: 300 (typography.label-sm)
    • Evidence: - Font: 14px, weight 400 (Buttons)
    • Finding: No line contains low confidence; component typography set by rule is seen because label-sm resolves to 10px/300 while the button prose reports 14px/400. Type-order, dropped-heading and transparent-shadow-layer issues are absent; no brandmd JSON, CSS or Tailwind output was supplied for comparison, and the generator release cannot be inferred.

    Low confidence

    none

    Generator issues

    1. Type levels out of size order: not present in this file. The body-md 16px and body-sm 14px names agree with their order.
    2. The largest heading dropped from the frontmatter: not present in this file. display is 48px, matching the largest Type scale heading.
    3. Fully transparent layers kept in shadow values: not present in this file. Every captured shadow layer has positive alpha.
    4. Names by position in brandmd's CSS and Tailwind outputs: not checked: no CSS or Tailwind output given.
    5. Component typography set by rule: seen in this file. components.button-primary.typography references typography.label-sm (10px, weight 300), while Buttons, Font says 14px, weight 400; use the prose value recorded in the sheet.

    Known fixes recorded by this skill: shadow placeholders were fixed in 0.17.1; heading and body naming, CSS and Tailwind radius mapping, and Tailwind spacing were fixed in 0.18.0. Shadow names still follow frequency, and component typography remains chosen by rule. DESIGN.md does not record which release wrote it.

    The paid skills read this file first and check its source line against DESIGN.md.

    agent with this skill installed, on a project whose DESIGN.md

    Connects securely to your tools. The creator never sees your data.

    About this skill

    No extractor can know which colour is your real call to action, whether a grey is a surface or a button, or what your font is called outside CSS.

    This skill runs 13 checks over the file, from the call to action and font names to spacing, shadows and dark data. It asks only what the file cannot answer, in one message, most questions with a default. Then it writes DESIGN.review.md next to DESIGN.md: your answers at the top, each with the DESIGN.md lines it rests on, a token sheet that gives every value its source and a status such as observed, derived, assumed or missing, and one block per check quoting the lines it rests on. DESIGN.md is never edited.

    It works with any Google-spec DESIGN.md. One way to get the file is brandmd, which extracts it from a live site.

    Free cloud run: paste your DESIGN.md, plus an earlier review or the extractor's JSON output if you have one. It asks nothing and writes nothing to disk. You get the whole DESIGN.review.md in one block, every question on its default, then an offer to print it again with your answers, and a short report. The pasted dry run printed a 59-row sheet, all 13 check blocks and four questions on their defaults.

    Use when you want to know which values in a DESIGN.md to trust before you build on it.

    Known limitations

    • It reads text only. It does not open the live site, so it cannot tell which of two candidate colours the site really uses. It lists each one with where it came from, and the call is yours.
    • No contrast ratios, and no colour conversion beyond reading the alpha of an 8-digit hex. An input border written as rgba() or oklch() becomes a question. Shadow colours in oklab() or oklch() stay as written.
    • Missing values are never filled in. A file with no surface or border colour gets missing rows, and later skills keep your existing values for those roles.
    • Dark data gets a verdict for you to read: none, real, not real, dark-first or unclear. Dark values from the prose are marked unverified.
    • Shadows are recorded as computed styles, split into layers and named. Current output removes recognized zero-alpha layers; the cleanup still handles older files that keep them. A search of your source code will not find them as written.
    • In a brandmd file it names the known brandmd issues it sees, and can name the known fixes, but it cannot say which brandmd release wrote the file.
    • A capture of a block page, a login page or another site fails the gate. With no earlier review it writes a short one that says to extract again from a normal page of the site. An earlier review is left as it was, answers included. Either way it stops there.
    • Older brandmd files, with numbered headings and no frontmatter, get only the checks whose labels match.
    • In the pasted run it sees only what you paste: DESIGN.md plus at most one more file, an earlier DESIGN.review.md or the extractor's JSON output. It finds nothing else and writes nothing.
    • It edits no file except its own DESIGN.review.md.

    • It reads text only. It does not open the live site, so it cannot tell which of two candidate colours the site really uses. It lists each one with where it came from, and the call is yours.
    • No contrast ratios, and no colour conversion beyond reading the alpha of an 8-digit hex. An input border written as rgba() or oklch() becomes a question. Shadow colours in oklab() or oklch() stay as written.
    • Missing values are never filled in. A file with no surface or border colour gets missing rows, and later skills keep your existing values for those roles.
    • Dark data gets a verdict for you to read: none, real, not real, dark-first or unclear. Dark values from the prose are marked unverified.
    • Shadows are recorded as computed styles, split into layers and named. Current output removes recognized zero-alpha layers; the cleanup still handles older files that keep them. A search of your source code will not find them as written.
    • In a brandmd file it names the known brandmd issues it sees, and can name the known fixes, but it cannot say which brandmd release wrote the file.
    • A capture of a block page, a login page or another site fails the gate. With no earlier review it writes a short one that says to extract again from a normal page of the site. An earlier review is left as it was, answers included. Either way it stops there.
    • Older brandmd files, with numbered headings and no frontmatter, get only the checks whose labels match.
    • In the pasted run it sees only what you paste: DESIGN.md plus at most one more file, an earlier DESIGN.review.md or the extractor's JSON output. It finds nothing else and writes nothing.
    • It edits no file except its own DESIGN.review.md.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 14 days ago

    • Free with an Agensi account

    Permissions

    Read Files
    Write Files

    Reads DESIGN.md, an earlier DESIGN.review.md, and brandmd's JSON, CSS or Tailwind output when you give it. Writes one file, DESIGN.review.md. Never edits DESIGN.md or any other file, never runs a command, never uses the network.

    Listed14 days ago

    What's inside

    Frequently Asked Questions

    No. It writes one file, DESIGN.review.md, and the only existing file it edits is its own earlier review. It writes no corrected copy either. brandmd has a drift check, and a corrected copy breaks it: the next extraction would report every corrected colour as repainted. So the corrections live in the review, as answers and statuses, and DESIGN.md stays as it was extracted.

    No. Any file in the Google design.md format works, hand-written ones included. Some checks look for known gaps in brandmd files, such as a text colour it made up or type levels out of size order in older files. In any other file those tests are skipped and the values the author wrote count as observed. A check whose lines are not in the file says "not applicable" and names the missing line. Tokens with other names get their own rows at the end of the sheet, and it asks which token plays which role, with a default only where the file's prose names the role. If no DESIGN.md is found, it says brandmd can extract one from a live site, or that you can write one by hand, and stops without writing anything.

    Only what the file cannot answer, all in one message, each question with its candidates and where each came from. Three are asked every time, unless an earlier review already holds your confirmed answer: which colour is the call to action, what the font is really called and what it falls back to, and whether the spacing scale is the one to use. The brand name is asked for every brandmd file, whose name is the page title, and for any file with no name. Up to six more come only when a check raises them, such as whether a grey is a real surface or whether the site is light or dark. Most have a default. A few do not, and their rows stay missing until you answer. Answer any of them, or reply "defaults" to keep every default. Your answers are marked confirmed and the defaults assumed, so anyone reading the file later can tell them apart. In the pasted run it asks nothing, and every open question stays on its default.

    It reads the DESIGN.review.md it wrote last time. An answer you confirmed is reused while the lines it rests on still read the same in DESIGN.md. If you extracted again and those lines changed, the question comes back with the old answer as a candidate. Questions left on their defaults are asked again. The file is replaced only between its start and end markers, so notes you wrote outside them stay. If the markers are broken or missing, it asks before writing. If the new capture fails the gate, the old review is left as it is.

    You do, first. It tells you what in your DESIGN.md was seen on the page, what was guessed and what is missing. The other skills in this set read its call to action, font and spacing answers, and the theme-wiring skill also takes the surface answer. They skip those questions while the review still matches your DESIGN.md, and run their own checks for everything else, dark data included. You do not need any of them to use this one.

    Yes. What the skill writes for you (DESIGN.review.md) is yours to use anywhere, client and commercial work included. The skill files themselves are for your own use under the marketplace's terms; please don't repost or resell them.

    New in Frontend & Web Apps