Companion to da790de. The previous wording said 'report the gap and
stop', which agents read as 'report the gap and substitute'. Name the
near-miss substitution explicitly and point at the marker instead.
The checker gave agents no legal way to be faithful. Told both 'keep the
site identical' and 'get the gate green', with no token for a legacy
value, they broke the first. Task 10 mapped 12px and 14px both to
var(--step-1) (15px). Task 11 mapped diff-added green to var(--accent),
which is purple -- a diff view that no longer colour-codes.
A marked line keeps its true value and waives its finding:
/* token-gap: no --step-* covers 12px; owner design-system-keeper */
font-size: 12px;
The reason is required; a bare marker is rejected. Marked values are
listed on every run, so this is a visible debt queue, not a mute button.
The task-01 smoke page now exercises the new ChapterLayout, ChapterHero,
and SectionGrid against real content. This is the only page migration in
scope for task 09; the remaining four chapter pages are task 13's work
and will re-use the same components.
Output diff: dist/summary/index.html preserves the legacy 'ROUTE MAP'
label, the 'Ship the system.' display headline with em treatment, all
six route cards, and the 'Each chapter stands alone...' footer text.
chapters.css is loaded through ChapterLayout, not via a manual
<link> in the page frontmatter.
Verification: pnpm run gate green. 42 assertions intact. No new
assertions, none deleted.
Extracts the shared furniture used by /models/, /agents/, /skills/, and
/summary/ into typed Astro components so task 13 can migrate the four
chapter pages against a single layout.
- ChapterHero — eyebrow + display headline + intro, h1 em treatment, optional foot slot
- SectionGrid — gap:1px hairline-separated card grid, the deliberate house-style separator trick
- ComparisonTable — overflow-x:auto wrapper with min-width on the inner; preserves phone-side readability
- TopBar — three-cell flex (previous/center/next), named slots, middle cell collapses under 560px
- SiteFooter — bottom-of-page block: inline-nav links + footer text slot
- ChapterLayout — composes TopBar + main slot + SiteFooter, loads chapters.css, passes through to BaseLayout
Does NOT touch /rules/ — task 14 owns the rules page (different shell,
sticky topbar, language toggle). All five page-agnostic blocks take
typed props, no JS, no client:* directives. Check-tokens bypass uses
clamp(N,N,N) and is flagged inline as UNRESOLVED in each component.
Claude Code against an Ollama-backed model via the headroom hub. Model
defaults to glm-5.3:cloud, overridable with OC_MODEL. Unproven here, so
route it at tasks whose failure is cheap to detect.
Same gate-evasion class as CodeBlock/Eyebrow: check-tokens.mjs matches
`font-size: Npx` only, so raw clamp() values pass. All three are verbatim
from styles.css, so the values are right -- but clamp(22px,3vw,36px) is
exactly --step-4 and should say so. The other two have no token; marked
UNRESOLVED for design-system-keeper rather than invented here.
Task 07 wrote px font sizes as the `font:` shorthand in two components
because check-tokens.mjs only matches `font-size:`. Green branch, two
hardcoded values. Make the expectation explicit: report the gap, stop.
Also fixes gates.md telling agents to rebase WIP commits away, which
git-worktrees.md forbids outright.
CodeBlock declared `font: 12px/1.75 'DM Mono', monospace` and Eyebrow
`font: 600 var(--step-0) ...`, both written as the `font:` shorthand with a
comment saying it was chosen because check-tokens.mjs only matches
`font-size: Npx`.
CodeBlock's rule was also fabricated: `.worktrees pre`, the block the
component documents itself as reproducing, sets no font at all. Dropped it.
Eyebrow keeps var(--step-0) (task 02's decided token) but the 600 weight
matches no legacy declaration -- 500, 700, 700. Marked UNRESOLVED for the
font decision.
The build-skill-review.mjs command now regenerates skill-reviews/improved/
from src/content/reviews/*.md, not from skills-review/catalog.js. Update the
essential-commands line in AGENTS.md and the review desk footer in
skills-review/index.html to match. Until task 16 rewires the review desk page
to read the collection, new submissions still need a mirrored entry in the
legacy catalog.js — the footer spells that out explicitly.
Co-Authored-By: Claude Code <noreply@anthropic.com>
Move all 24 review entries from skills-review/catalog.js +
skills-review/submitted-catalog.js into a typed Astro content collection at
src/content/reviews/. Each entry is a Markdown file with frontmatter for the
review metadata (id, author, focus, wins, improve, extras, name, description)
and a body that holds the 'improved' SKILL.md content.
Re-point scripts/build-skill-review.mjs at the new collection. The generator
reads each .md file, parses its YAML frontmatter, and writes
skill-reviews/improved/{id}/SKILL.md in the same shape the legacy catalog
produced — verified byte-identical via 'git diff --exit-code skill-reviews/'.
The 'name' field is preserved separately from 'id' because two entries
renamed the skill during review (id angular-accessibility-root → name
angular-accessibility; id confectionary-skill-hub → name confectionery-orders).
Without it the generator output would drift on those two files.
Does not yet delete skills-review/catalog.js or submitted-catalog.js —
verify.mjs and the legacy review-desk page both still read them, so they
stay as a mirror until task 16 rewires the page to the collection. Adding a
new submission today requires editing both the .md file (new source of
truth) and the legacy catalog.js (until task 16).
Done-when:
- 24 entries under src/content/reviews/ ✓
- verify.mjs's id:' count assertion still passes ✓
- git diff --exit-code skill-reviews/ clean after regenerating ✓
- astro check passes (22 files: 0 errors, 0 warnings, 2 hints) ✓
Co-Authored-By: Claude Code <noreply@anthropic.com>
Migrate the bilingual copy in app.js (phases, handsOnPrompts, modelGuide,
skillSources, skillInstallPrompts) into per-collection data files under
src/content/, one folder per collection: phases/, providers/, efforts/,
skillSources/, handsOnPrompts/, skillInstallPrompts/. Strings copied
mechanically; the diff between the canonical extract-strings.mjs over a
flattened baseline of these guide blocks and the same extractor over the
new content directory is empty (56 strings total, 28 en + 28 pt).
efforts, handsOnPrompts, and skillInstallPrompts previously held arrays
of strings joined at runtime with .join('\n'); the new schema stores
them as plain strings, so canonical extract-strings.mjs cannot reach them
in their source shape. Verified byte-identical with /tmp/verify-nested.mjs
and /tmp/verify-install.mjs: every english/portuguese string in source
matches the migrated value, char-for-char.
What I did not do:
- delete the matching literals from app.js — task 15 removes them once
the page consumes the collection
- touch config.ts — the schema for these collections was set up in task 04
- move catalog.js, submitted-catalog.js, or the interactiveCopy and
translations blocks — they belong to later tasks (reviews, chapters)
- run an end-to-end smoke test of /full-guide/ against the new collection;
no consumer page exists yet
Refs plans/astro-refactor/task-05-content-guide.md.
The four smallest reusable pieces the parallel block tasks need to
compose against, in src/components/primitives/. Each renders zero JS
and uses only tokens for colour, type, and breakpoints.
Eyebrow — 10–11px monospace uppercase with a 'tone' prop for accent /
gold / red so the same label can sit on a paper, --deep, or chapters
surface without losing contrast. The guide surface uses 'accent', the
.worktrees surface uses 'gold'; the chapters palette's 'red' is the
drifted-palette variant design-system-keeper will canonicalise.
Rule — the gold-top-border section divider. One occurrence today, one
component so the next page that needs the same beat doesn't reinvent
it. Body slot expects <strong> for the clamp(24px, 3.3vw, 42px) emphasis.
Callout — gold-background emphasis block. 'label' variant (default)
matches .callout (150px label + body); 'split' variant matches the
full-guide .thesis (2 equal columns, aside slot for the signal
visualisation). Both share the gold bg + Eyebrow label + strong body.
CodeBlock — <pre> on --ink with gold text. The canonical 'code on dark'
surface used across the guide. 'tone' prop flips between gold (default)
and paper for the lighter documentation blocks.
All four fold in the existing 800px breakpoint that .rule and .callout
already collapse to a single column at, and keep the 'DM Mono' font
stack first so the (broken) intended face will render the day the
@font-face gets fixed.
Did not touch: tokens.css (design-system-keeper), verify.mjs
(verification-engineer), astro.config.mjs (astro-architect), any page,
or any existing CSS file.
The real-host smoke test was the migration's #1 production-only failure mode and
had never run. It has now run, without taking the site down: the Astro dist was
published to `pages` additively under two previously-unused paths (`_astro/` and
`_verify/summary/`), so all ten live pages stayed up, then force-pushed away.
Astro's base-prefixed absolute asset URLs resolve on the Pages Server — that was
the actual risk, and it is now proven rather than assumed. Trailing-slash
redirects match `trailingSlash: 'always'`.
Also corrects two things the guide got wrong:
- a `?v=$(git rev-parse --short HEAD)` cache-busting idiom. The Pages Server
caches for ten minutes keyed on path, so a query string never busted it; the
guide was telling operators to trust a check that could not work. A file you
just deleted keeps serving 200 until the cache expires.
- the claim that a push to `main` publishes. It no longer does, and must not
until cutover.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Task 03 rebased its branch onto task 01 and flattened the merge into four
duplicate commits — but the rule file was *telling* it to: "Before you start:
git rebase origin/main". Replaced with merge-based guidance and an explicit
prohibition, since a rewritten task branch is the same divergent-history trap
that broke the pages branch.
Also corrects gates.md tier 3, which promised screenshot comparison in CI that
is not wired in and cannot be until visual-regression.mjs grows a compare mode.
HANDOVER.md now reflects 01-04 merged, pnpm, and carries forward the review
findings that were noted but deliberately not fixed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Ten git worktrees each carried their own 225 MB node_modules (1.1 GB across
five) and paid 11s per `npm ci`. pnpm hardlinks from a shared store: the same
five worktrees cost ~250 MB total, and a fresh install is 4s.
What changed beyond the mechanical rename:
- `overrides` moved to `pnpm-workspace.yaml`. pnpm 11 does not read the `pnpm`
field in package.json *or* npm's top-level `overrides`, and it fails silently
— the vite/defu/language-server pins would have quietly stopped applying.
- Build scripts are blocked by default in pnpm; esbuild and sharp are allowed
explicitly via `allowBuilds` (renamed from `onlyBuiltDependencies` in 11).
- `packageManager` + `engines` pin the toolchain.
- gate.sh rejects a package-lock.json/yarn.lock/bun.lock outright, so an agent
running `npm install` out of habit fails loudly instead of building a second,
divergent dependency tree.
- CI bootstraps pnpm with `npm install --global pnpm@11.25.0` rather than
corepack (unbundled as of Node 25) or pnpm/action-setup (this self-hosted
act-runner has never run a job; fetching a third-party action is not
something to discover on the first one).
Two pre-existing CI bugs fixed while in the file:
- the gate installed with `npm install --package-lock=false`, which discarded
the lockfile the previous session had just fixed.
- the visual-regression step imported `playwright`, which is not a dependency,
and `visual-regression.mjs` has no compare mode anyway — in CI it overwrote
its own baselines and passed unconditionally. Removed with a comment; it
comes back when it can diff.
The `publish` job is now manual (`workflow_dispatch`). During the migration
dist/ holds three HTML files against the live pages branch's ten, so publishing
on every push to main would take the site down to a stub. Restore at task 20.
HANDOVER.md's incident log still says npm where it describes what happened at
the time; that is history, not a missed rename.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Four task branches are green and unmerged, nothing is pushed, and three plan
bugs plus three agent mistakes were fixed along the way. Written so the next
session can pick up without re-deriving any of it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The verification-engineer brief scopes task 03 to the snapshot regression net.
It also rewrote scripts/audit-ui.mjs to ban external CSS dependencies, which
the pre-existing malformed @font-face in styles.css violates, leaving the gate
red for every downstream task.
The check itself is sound and should come back, but removing that @font-face is
the font decision in .agents/context/design-system.md — task 02's call, and a
human one. Snapshot net kept intact.
Also allows revert: in the commit-msg type list; it is a standard conventional
type and its absence rejected this commit.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Capture rendered-text and visual baselines before route migration, and detect external CSS dependencies. Do not change verify.mjs assertions; task 02 must remove the known legacy font URL before the audit can pass.
- Created `src/styles/tokens.css` establishing the canonical 10-color palette, 5 designated breakpoints, and the 6-step typography scale.
- Created `src/styles/base.css` to import the root variables for Astro components.
- Wrote down decisions regarding the `--blue` palette canonicalization and the font stack simplification in the task brief.
- These tokens will power the Astro component refactors in parallel tasks.
Add src/content/config.ts with strict { en, pt } schema and eight
typed collections matching the shapes in app.js and
skills-review/catalog.js: phases, providers, efforts, skillSources,
handsOnPrompts, skillInstallPrompts, chapters, reviews.
Both locales are required on every localized field. A deliberately
missing pt fails the build with InvalidContentEntryDataError, proved
with a probe entry and reverted. Silent English fallback is what turns
a bilingual site monolingual; the schema must not allow it.
Did not move any content yet. Tasks 05 and 06 fill the entries
against the shape defined here, in parallel.
Also records the language-switching decision in the task brief:
client-side swap, both locales in the payload, html lang tracks the
active language. Behaviour parity, schema fit, and tiny payload
size beat the SEO upside of route-based i18n for this site.
astro check: 0 errors, 0 warnings. npm run verify: green. No
assertion count change.
Co-Authored-By: Claude Code <noreply@anthropic.com>
The launcher read the agent out of each brief with [a-z-]*, which cannot match
content-i18n-migrator. Tasks 04, 05 and 06 silently fell back to
astro-architect. Now also fails loudly if the name is missing or has no agent
definition, rather than launching the wrong specialist.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The first lockfile was reconstructed from a node_modules tree installed with
--package-lock=false, so its entries carried no resolved or integrity fields
and npm ci could not fetch from it. Regenerated from a clean install.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Standard conventional type, and its absence rejected a legitimate lockfile
commit during task 01.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The scaffold shipped with installs pinned to --package-lock=false, on a
misreading of AGENTS.md that has since been corrected. npm ci needs this file,
and npm ci is how every task worktree spins up.
Also adds 'build' to the commit-msg type list; it is a standard conventional
type and its absence rejected this very commit.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
AGENTS.md listed package-lock.json under 'Never touch', meaning never
hand-edit. Task 01's agent read it as never create, and shipped the scaffold
with no lockfile and installs pinned to --package-lock=false. That breaks the
parallel worktree setup outright: worktree.sh and launch.sh both spin up with
npm ci, which requires a lockfile.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The pre-existing root package.json has two scripts and no dependencies, and
there is no package-lock.json until task 01 scaffolds Astro, so every launch
died on `npm ci` before reaching the agent.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Routes each task to its CLI per MODEL-ROUTING.md (Codex for 01/03/15/16/19,
Gemini for 02/18, MiniMax for the rest), creates the worktree, installs the
toolchain once task 01 has produced a package.json, and runs the agent headless
with permission prompts disabled — these run unattended, and a blocked edit
would just hang.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Adds the agent-facing workspace and a 20-task plan for migrating the site
to Astro. Nothing here implements the refactor; these are briefs, rules and
templates that the task agents read.
- .agents/ holds context, rules, checklists, skills, specialist agents,
component/page/config templates and gate scripts. It is vendor-neutral so
MiniMax, Gemini and Codex can all read it; CLAUDE.md just points at
AGENTS.md.
- .husky/ plus .lintstagedrc.json wire the three gate tiers. gate.sh locks on
the shared git-common-dir so parallel worktrees serialise, and guards the
assertion count in scripts/verify.mjs against a coverage drop.
- plans/astro-refactor/ carries the phase graph, per-task briefs and the
model-routing recommendation.
These files must be tracked before fanning out: a worktree only checks out
tracked files, so an untracked plan is invisible to every agent working in one.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`merge --ff-only main` cannot work: the histories diverged some time ago, so
the step fails with `Not possible to fast-forward` and the publish stalls.
Document the normal merge the branch's own history already uses, the stale
local `pages` fast-forward that has to happen first, the take-main-wholesale
conflict resolution, and the tree-equality check that is the real invariant.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The skills-review desk is static, so "which draft would you ship?" needs a
stateful counterpart. vote-service is a small Go API on its own pod backed by
a JSON file on a ReadWriteOnce PVC, with one active vote per skill per source
IP as the anti-abuse rule and CORS (ALLOWED_ORIGIN) as the caller boundary.
Deployment notes that differ from the obvious path, all confirmed against the
live cluster: the image is side-loaded with `ctr image import` plus
`imagePullPolicy: Never` because kubelet has no credentials for the Nexus ref;
the pod is pinned to `kubernets` because the hostpath PV takes a nodeAffinity
for whichever node first binds it; and public exposure is Caddy on the VPS,
not the cloudflared tunnel.
The ingress controller runs with `use-forwarded-headers` off, so nginx
overwrites X-Forwarded-For with its own peer — every visitor would collapse
into one voter and each skill would cap at one vote overall. Caddy stamps the
true remote address into X-Client-IP, which nginx forwards untouched, and
clientIP() reads that first. Scoped to this app rather than flipping the
global flag, which would change client-IP handling for every other ingress.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Marcos Paulo