[samuv]

Summary

What lands

• BFF: fourteen /api/v1 routes projecting configured skills, versioned learned skills (detail, unified diff, lifecycle history, activate / archive / reject / rollback carrying the daemon's expectedRevision), the learning-proposal queue (decide, undo promotion), session reflection receipts, and the user model with its daemon-curated consolidation plans; each surface gated live on its own capability; a missing memory key is a 404.
• Web: knowledge workspace (configured and learned views, learned-skill detail with diff, learning review, memory consolidation dialog); "Skills" joins the nav.

Verification

81 server and 214 web Vitest proofs (15 ACs, 0 missing); task studio:check, task docs pass; smoke against mecated --mock, which advertises every knowledge capability (inventories, memory, plan generation exercised live).

Further amendments

Implements the amended knowledge contract (see the plan layer below):

• The server offers actions by lifecycle state and refuses a rollback without a target.

• The stale-revision proof uses the daemon's real proposal_conflict code.

• The detail page recovers the old and new text from the daemon's diff and renders a line diff.

• Plan status. This layer's diff moves its acceptance plan to landed (header and index), per the plan lifecycle; the transition is authoritative only when this PR merges, and task ac-trace-strict now gates the plan.

Re-verified on the rebased stack: task studio:check passes (120 server + 254 web Vitest), and task ac-trace-strict passes with every Studio plan gated as landed (141 ACs, 0 failures).

Development stage

• Plan / Interface
• Implementation
• Combined
• Spike / Routine / Cleanup

Contract linkage

• Work classification: Bounded
• Classification rationale: adds the knowledge product surface (skills, learned skills, proposals, reflection, user memory) inside the boundary ADR 0351 fixed
• Decision record: None — see the plan layer
• Human waiver of spine: Yes — the directing human chose stacked (gh stack) delivery for the whole Studio series: nothing merges until every layer is reviewed, so the plan's approval-by-merge event is deferred to the end (recorded in each plan's Delivery line), and each implementation layer records the exact plan commit it built against.
• Acceptance plan: docs/acceptance/studio-knowledge.md
• Human decisions resolved and recorded: Yes
• Plan / Interface PR: docs(plan): Mecatl Studio knowledge acceptance plan #1759
• Approved commit baseline: 429f6485db43306f39f2b01f4ff6e0df2e74309e (the plan layer below; not yet merged — stacking exception above)
• Combined/exemption rationale: N/A

Interface conformance

• Matches the plan's contract.

Issue relationship

Relates to #1749 (parent #1736; stack #1766). Non-closing: the step completes when both of its layers merge.

Type of change

• Behavioral/interface plan
• Bug fix
• New feature
• Refactoring (no behavior change)
• Dependency update
• Documentation/process

Test plan

Baseline checks

• Acceptance-plan checker
• Linting (task lint)
• Offline test suite (task test)
• Offline demo (go run ./cmd/mecademo)
• Markdown changed: docs generation/link checks (task docs)

Studio gates (this workspace is not covered by task lint / task test, which are Go-only): task studio:check (Biome, typecheck, Vitest, generated-drift), task ac-trace — all pass; see Verification above.

User-facing change

Studio gains the Skills page: configured and learned skills, learning review, memory consolidation.

Special notes for reviewers

Part of a fourteen-layer stack; review bottom-up. Nothing merges until the whole series is reviewed.