Lock

X2 Lock

Copy the approved design artifact down into the repo, verify it is complete, stamp its provenance, and request the lock. The lock is the acceptance contract for the slice; everything the builder ships is held to it. The pilot ran without this step and paid twice: the design system was never imported so the client's tokens were scraped from what happened to be readable inline (MET-04), and the prototype was treated as locked for days before anyone asked for the approval (turn 25).

The import

Pull the complete artifact set from the design project into design/prototype/ (the canonical home, INV-01):

• the prototype source (<name>.dc.html),
• the complete design-system export under _ds/ (the token css, component css, fonts),
• the supporting scripts.

The sanctioned route is the on-disk export: the owner places the design project's complete export in the repo (a handoff folder under design/), the lock copies out what it needs and deletes the handoff. A context-carried import (the design MCP) is the fallback for small artifact sets; it cannot carry binaries, and the third project moved to the on-disk route after channel limits bit. Either way, verify the copied tree against its source: the file inventory matches and no copy artifacts exist (a duplicated or nested directory survived the third project's import verification and was caught only by the owner's eye).

Never derive tokens from values readable inline in the prototype source. The pilot did, and silently produced a partial, light-only, placeholder token set with the coverage floor skipped. The client's token source derives from the export on disk, nothing else.

Done-checks (all must pass before requesting the lock)

• The token file on disk defines the full variable set: the more-than-fifty floor is met, not skipped.
• Both themes are present: the dark root and the light override.
• Fonts resolve by a named path: binaries checked in, or a self-hosted font package pinned under the dependency ledger (the pilot's answer; the design MCP cannot carry binaries through context).
• design/prototype/README.md records provenance: source design-project URL, import date, the portion being locked. On approval, it gains the lock stamp.
• design/ledger/ exists for the per-slice fidelity ledgers that derivation will add.
• Where a design brief exists for the portion: the copied-down screens and their navigation map onto the brief's screen inventory and action edges, both directions. An orphan on either side (a brief surface the prototype lacks, a prototype surface the brief never named) is surfaced at the lock for a ruling, never silently absorbed.
• The exposure review. A prototype can carry security-shaped choices, and transcription would ship them. Walk the portion for: entry points it exposes (an admin door on a public page), identifiers it addresses things by (sequential versus opaque on any public surface), and embedded exposure values (domains, addresses, ids). Each finding is surfaced for a ruling at the lock, never silently transcribed; the second project shipped and then moved both instances it missed (record: NS-3).

Human-turn contract

• Requests gate 2b: the lock (MET-07): the copied-down artifact is approved, explicitly, now. This is UI-4's chargeable turn. A portion is not locked because work moved on; it is locked because approval is on record. Request it at this moment, not retroactively.
• The request is put in plain language, the owner's own words; method vocabulary rides alongside, never instead (P3).
• A re-lock after a spine revision is the same turn again for the affected portions.

Questions this skill used to carry, now ruled (invariants pass, 2026-07-11)

• The locked artifact is the DC source alone. No static render is checked in; a render, if ever produced, is viewing convenience and never authoritative. Approval happens where the prototype renders (gate 2a, Claude Design); the lock is the copy-down diff (INV-01, ruled).
• The design system lives under design/prototype/_ds/, provenance-true where the import lands. A shared house-system home is deferred with a trigger: the second product reusing the same system makes it a pinned, hash-ledgered vendored asset under DEP-1, not a shared directory.
• The import is the COMPLETE export (the .dc.html plus the full _ds/, both themes, more than fifty variables), per the manifest; never values scraped from the .dc.html (MET-04, ruled).

What this skill must NOT produce

• No tests, no fidelity ledgers, no code. Derivation starts only after the lock is on record.
• No edits to the prototype; a change goes back through design and re-approval.

Next

Gate 2b on record, then derive-tests.