| name | landing-page-convergence |
| description | [Tier 2 · moderate autonomy · full review gate] Force a production landing page that has DIVERGED from an approved design into full fidelity with that design - section by section, with a named divergence checklist as the forcing function so it converges instead of drifting. Use when a live page no longer matches its design export and needs to be brought back into line (a single page/site, not a whole app - that's design-integration). Extracts the design system, diffs production against it, and closes each divergence as its own PR. Runs via the autonomous-fleet-core engine. Trigger on: "make the landing page match the design", "the production page diverged from the design", "bring the landing page into parity", "fix the landing page to match the mockup". |
| license | MIT |
| compatibility | Requires git and gh CLI in the target repository |
| metadata | {"author":"ravidsrk","version":"1.0.0","tier":"2","fleet-component":"mission"} |
| status | exploratory |
Status: exploratory. This mission is documented but has not yet been run end-to-end on an external repo with an archived run-archive. It is preserved for future promotion (see docs/exploratory/missions/README.md § Promotion criteria). Do not invoke this skill in a production fleet until it has been promoted.
Mission: landing-page-convergence
Required skills
Before executing, activate these skills and read their full instructions:
autonomous-fleet-core — read references/engine.md and references/composition.md when coordinating- One runtime adapter:
autonomous-fleet-adapter-orca, autonomous-fleet-adapter-claude-code, autonomous-fleet-adapter-grok, or autonomous-fleet-adapter-codex
Follow the core and your adapter in full, then apply the mission parameters below.
Do not load a second mission skill in the same run. For chained missions, use fleet-program.
Optional skills
| Skill | Activate when | If unavailable |
|---|
qa-only | Production/staging URL; report-only before final section PR | Screenshot + @claude review only |
browse | Coordinator spot-checks live page between sections | Screenshot + manual diff |
Community catalog: autonomous-fleet-core → references/community-skills.md.
Worker skills
| Role | Skills | If unavailable |
|---|
| @grok (build sections) | frontend-design, frontend-ui-engineering | Design extract values exactly |
| @grok (section QA) | qa when URL available and fix-verify loop requested | Fidelity gate per section |
| @claude (extract/diff, fresh review, integrator) | — | Mission fidelity gate |
Deferred missions
Record in final readiness doc under Recommended next missions and in DECISIONS.md.
| Finding type | Route to |
|---|
| Whole app must match design | design-integration |
| Copy/positioning strategy undefined | take-product-to-completion (boundary pass) |
Empirical note: No matching task category in arXiv 2601.15195 - full review gate. The thing that
makes it converge rather than drift is a NAMED DIVERGENCE CHECKLIST that is part of the
termination condition: the run isn't done until every listed divergence is closed, and each PR
declares which divergence it closes.
DESIGN SOURCE (the user supplies one)
A Claude Design export/URL or design file. A worker fetches/opens it and READS ITS README first,
then extracts the EXACT design system. This export's values are authoritative - match precisely,
don't approximate.
DIVERGENCE CHECKLIST (the forcing function)
If the user names specific divergences, transcribe them as D1..Dn - these are the floor. T2 (the
production diff) confirms/details each AND surfaces any additional divergences (add as new
D-items). Each D-item is a FIX the build must close and the reviewer must verify against the
design export. The run terminates only when every D-item is CLOSED.
GOAL
Bring the production landing page to full fidelity with the design export - same composition, type
scale, spacing rhythm, sections, component treatment. The design is the TARGET; production
changes. Converge, do not reinterpret or "improve on" the design. Keep existing copy and working
links/CTAs unless the design's section structure requires content the page lacks (then build it to
match the design).
ROLE PIPELINE
Builder choice (Stage-9 carryover). This mission keeps @grok as builder
because Stage 6 (prompts.md L1488, L2038) showed @grok+frontend-design
produces materially better visual fidelity on landing-page and design-parity
runs than @codex. Stage-9's general retirement of @grok (prompts.md L3013)
does not apply to design/landing missions.
- @claude EXTRACTS the design + DIFFS production (the frozen DIVERGENCE CHECKLIST).
- @grok rebuilds each section.
- A fresh build-blind @claude REVIEWS each PR: matches the design export
(tokens/spacing/type/layout), the claimed D-item is actually closed, responsive, no
placeholders.
- @claude is the INTEGRATOR: opens PR, conflict-aware merge, worktree cleanup.
LEDGER
<LEDGER_DIR>/landing-progress.md. Per-task flags: BUILT=<t/f> PR_OPEN=<t/f> REVIEWED=<t/f> MERGED=<t/f>. Plus the DIVERGENCE CHECKLIST D1..Dn, each OPEN | CLOSED via PR#n.
VISUAL BASELINE (before/after gate)
T2 records the exact baseline capture commands/queries that produced the before-state
screenshots in docs/landing-diff.md and <LEDGER_DIR>/landing-progress.md: command line or browser
query, URL/route, viewport/device, auth/seed state, screenshot path, and covered section/D-item.
T-FINAL replays those commands/queries like-for-like after all PRs merge, captures after-state
screenshots to parallel paths, and writes an explicit before/after comparison in
<LEDGER_DIR>/landing-readiness.md. Passing requires strict visual improvement toward the design for each
covered D-item. Screenshot existence alone is not evidence. If a baseline cannot be replayed, the
covered D-item stays OPEN until a replayable baseline exists and passes.
TASK STRUCTURE
- T1 DESIGN EXTRACTION [@claude] - fetch/open the design export (README first); extract the
EXACT system (tokens color/type/spacing/radii, component styles, type scale + weights, spacing
rhythm, section-by-section layout). Output docs/design-extract.md + place assets.
- T2 PRODUCTION DIFF [@claude, gated on T1] - diff the current production page section by
section against docs/design-extract.md; confirm and detail each known D-item with the exact
production files responsible; surface any additional divergences (add as D-items). Output
docs/landing-diff.md. Record VISUAL BASELINE capture commands/queries and write the full
checklist to the ledger.
- T-CONVERGE… [per section/D-item, loop] - each is one PR. @grok rebuilds that section to
MATCH the design exactly (correct tokens, spacing, type, layout, component treatment), closing
its D-item(s), responsive, real content → fresh build-blind @claude reviews (fidelity + D-item closed + responsive)
→ @claude merges. Do global-craft items (tokens/type scale/spacing rhythm/eyebrow labels) FIRST
as a foundation pass so per-section fixes inherit them. Parallelize non-overlapping sections;
serialize same-file. Update D-item close-status.
- T-FINAL [@claude] - build green, lint clean. Walk the page across mobile/tablet/desktop;
EVERY D-item CLOSED; matches docs/design-extract.md in tokens/spacing/type/layout/section
structure; fully responsive; no placeholders/dead links/console errors. Replay VISUAL BASELINE
commands/queries and include the strict before/after comparison. Output
<LEDGER_DIR>/landing-readiness.md
with fleet-outcome YAML (divergences_open), D-item summary, Recommended next
missions, all PRs. Ship as the final PR.
Runtime goal
After ledger init, SET_GOAL per autonomous-fleet-core/references/runtime-goals.md. Record
## Runtime goal in <LEDGER_DIR>/landing-progress.md. GOAL_COMPLETE only after ## DONE below.
Mission landing-page-convergence DONE: <LEDGER_DIR>/landing-progress.md all task flags true,
<LEDGER_DIR>/landing-readiness.md with fleet-outcome.status done and mission metrics satisfied,
./scripts/validate-fleet-outcome.sh passes, all PRs merged into BASE.
DONE
Every D-item CLOSED, every task BUILT=t PR_OPEN=t REVIEWED=t MERGED=t, page matches the design
export, replayed VISUAL BASELINE confirms strict improvement, <LEDGER_DIR>/landing-readiness.md exists.
Then send the FINAL report.
DECISION DEFAULTS (mission-specific)
- The design export is the TARGET; production changes to match. Converge, don't reinterpret or
"improve."
- The export's exact token/spacing/type values are authoritative; match precisely, don't
approximate.
- Keep existing copy and working links/CTAs UNLESS the design's structure requires content the
page lacks (then build it to match the design).
- Close global-craft items (tokens, type scale, spacing rhythm, eyebrow labels) early so
per-section fixes inherit them.
- Real content/assets; no placeholders on the shipping page. Mobile-first/responsive + a11y.
- Any ambiguity → the closest fidelity to the design export while converging.
