agentleFS
mxrsvmxrsv/spacevibe-deck/.claude/skills/landing-sync/SKILL.md

landing-sync

Skill41 starsChanged 23 days agoMITView on GitHub

Report which parts of the Deck landing page (marketing/landing-prototype, marketing/stage) are stale against the app (src/, electron/, docs, CHANGELOG). Read-only by default. Use for "landing drift", "sync landing", "is the landing stale", or as the "before release" check before tagging. Takes an optional git range.

  • Reads credentials

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read every new ## <version> section in full, not only the grep hits: a surface the landing sells can be rewritten without any of those …

    line 55
  2. Distinctive:

    Do not keep a hand-written table of what the landing mirrors.

    line 60
  3. Very specific:

    Every hit is Stale: the landing sells something the app no longer has.

    line 127

≈ “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

landing-sync — does the landing still describe the app?

The landing is hand-copied from the app. A few files import app code; everything else restates the app in prose, fixtures and comments that cite the app source they mirror. .github/workflows/landing.yml runs only on marketing/**, so a change under src/, electron/ or docs/ never fails a landing check. This skill is that check, run by hand.

Read-only by default. Produce the report, stop, and wait for the owner. Do not edit anything under marketing/ in the same pass — see "After the report". The disallowed-tools frontmatter blocks the edit tools while this skill runs, but Bash can still write, so the read-only rule still binds.

Shell state does not survive between tool calls, so step 0 writes the three values every later block needs to /tmp/landing-sync.env, and every block opens by sourcing it. Never retype a hash from memory; re-source the file.

0. Pick the revision

Arguments: $ARGUMENTS. Empty → judge origin/main from the last commit that touched the landing. <base> → that base, still judging origin/main. <base>..<rev> → both given. The production landing deploys from origin/main; the primary checkout can be far ahead of it and dirty, so the judged revision is checked out into a scratch worktree and nothing below reads the checkout you were started in.

REV=origin/main                                   # or the <rev> half of the argument
git fetch origin --quiet
W=/tmp/landing-sync-$(git rev-parse --short "$REV")
[ -d "$W" ] || git worktree add --detach --quiet "$W" "$REV"
cd "$W"
BASE=$(git log -1 --format=%H -- marketing/landing-prototype marketing/stage)   # or the <base> argument
printf 'REV=%s\nW=%s\nBASE=%s\n' "$(git rev-parse HEAD)" "$W" "$BASE" > /tmp/landing-sync.env
cat /tmp/landing-sync.env; git log -1 --format='%h %ad %s' --date=short "$BASE"

A landing commit resets the baseline only for what it touched. Drift older than the base is caught by the state passes in step 3 (c, d, e), which do not depend on the range, so run them every time.

1. What changed in the app since the base

. /tmp/landing-sync.env && cd "$W"
git diff --name-only "$BASE"..HEAD -- src electron docs/user docs/internals CHANGELOG.md
git diff "$BASE"..HEAD -- CHANGELOG.md | grep -E '^\+' | grep -iE '^\+## |retired|removed|replaced|no longer|renamed'

Read every new ## <version> section in full, not only the grep hits: a surface the landing sells can be rewritten without any of those words.

2. Derive the coupling map

Do not keep a hand-written table of what the landing mirrors. Derive it from the landing each run, in five passes. Each block starts from the scratch worktree.

a. Imports. The only live links; an app change here breaks the landing build or changes its pixels.

. /tmp/landing-sync.env && cd "$W"
grep -rnE "from ['\"][^'\"]*\.\./src/|new URL\(['\"][^'\"]*\.\./src/" marketing/landing-prototype marketing/stage --include='*.js' --include='*.mjs' --include='*.html' --exclude-dir=node_modules

b. Path citations. Comments naming the app file they copy. Classify each path: present at the repo root → coupling; present only under marketing/ → landing-internal, skip; present nowhere → the source moved or died, so the copy is unverifiable and goes to Probably stale until the fixture is checked by hand.

. /tmp/landing-sync.env && cd "$W"
grep -rnoE "(src|electron|docs)/[A-Za-z0-9_./-]+\.(ts|tsx|js|css|md|rs)" marketing/landing-prototype marketing/stage --include='*.js' --include='*.css' --include='*.html' --include='*.mjs' --exclude-dir=node_modules | sort -u

Cross each coupling path with step 1: a cited file in the diff → Probably stale, read the diff and the mirror together to promote or clear it.

c. Copy. marketing/landing-prototype/src/copy.js carries the panel text in every locale, so one wrong sentence appears once per locale.

. /tmp/landing-sync.env && cd "$W"
# Chords the landing promises. Each must map, in docs/user/keyboard-shortcuts.md,
# to the feature the sentence around it names.
grep -noE "⌘[⇧⌥⌃]*[A-Z0-9]" marketing/landing-prototype/src/copy.js | sort -t: -k2 -u
# Feature names come from the docs, not from a list kept here: page titles and
# shortcut actions. Grep the landing for each name that is a product noun.
{ sed -nE 's/^# (.+)$/\1/p' docs/internals/*.md docs/user/*.md
  grep -hoE '^\| [A-Z][^|]+' docs/user/keyboard-shortcuts.md | sed 's/^| //;s/ *$//'; } | sort -u

d. Version pins. Comments and fixtures that say which Deck they draw.

. /tmp/landing-sync.env && cd "$W"
grep -rnoE "Deck [0-9]+\.[0-9]+(\.[0-9]+)?" marketing/landing-prototype marketing/stage --include='*.js' --include='*.css' --exclude='*.test.js' --exclude-dir=node_modules | sort -u
node -p "require('./package.json').version"

A pin older than the current version is not drift by itself. It is drift when the surface it names changed since that version: check it against the CHANGELOG sections in between.

e. Retired subjects. The docs date every retirement. A page-level "is this page retired?" question misses a surface retired in the middle of a living page, so the unit here is the retirement line, not the page.

. /tmp/landing-sync.env && cd "$W"
grep -niE "(retired|removed|replaced|superseded)[^.]{0,40}([0-9]{4}-[0-9]{2}-[0-9]{2}|[0-9]+\.[0-9]+\.[0-9]+)" docs/internals/*.md docs/user/*.md

For every line, take the thing it retires — the noun phrase before the verb, plus the chord or control that went with it — and grep the landing for it and its plain variants. The subjects are read off the lines each run; today's lines give, for example:

. /tmp/landing-sync.env && cd "$W"
grep -rniE "agent board|⌘⇧O|checkout card|checkout is a card|fold(s|ed)? (its|the) agents|card.{0,20}\+" marketing/landing-prototype/src marketing/stage --include='*.js' --exclude='*.test.js'

Every hit is Stale: the landing sells something the app no longer has. Map a hit to its panel through PANELS in marketing/landing-prototype/src/tour/index.js and the copy key it uses, and the fix covers the scene and the key in every locale.

Also grep the landing for the stem of every app file in the step 1 diff (agent-rail, tab-manager, …). A hit is a mirror without a citation; treat it as a citation.

3. Report

One entry per landing surface, keyed file:line. Group as Stale (the app contradicts it), Probably stale (its source changed, moved or died and the mirror has not been re-read), OK (checked, still true). Each Stale or Probably stale entry carries:

  • App change: what happened, with the commit, CHANGELOG section or docs line.
  • Evidence: the landing line and the app line, both quoted.
  • Wrong now: what a reader is told that is no longer true.
  • Fix: the replacement copy, fixture or deletion. Tour panels come in pairs: the scene under marketing/landing-prototype/src/tour/scenes/ and its text in copy.js, in every locale.

End the report with REV and BASE from /tmp/landing-sync.env and which of the five passes ran. An empty Stale group is a finding only if every pass ran. Add two last lines:

  • Not checked: each surface or claim you could not judge, with the reason. Never omit it and never guess; write "Not checked: none" only when nothing was skipped.
  • Skill gaps: anything in this skill that was wrong, missing or cost a retry, as a suggested edit. Suggest only; do not edit the skill.

Then clean up:

. /tmp/landing-sync.env && git worktree remove --force "$W" && rm /tmp/landing-sync.env

After the report

Landing edits start only after the owner approves the report, and follow the landing's own workflow: a worktree, a feat(landing) or fix(landing) commit per surface, the landing.yml tests and npm run build:landing, and the owner's eye review of the rendered page before merge. The landing stays free of tests that pin app data (owner decision, DECK-85); this skill is the check that replaced them, so do not turn its passes into CI.

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.