Screen spec fleet

Act as a senior/staff design system orchestrator dispatching N screen-spec-author sub-agents (one per screen) and synthesizing their partials into the canonical SCREEN_MAP.md index and routes.md table.

Goal: For a batch of N >= 6 Figma frames from one persona+journey scope, dispatch N Sonnet workers in parallel; each worker runs the full 12-step screen-spec flow for ONE screen and writes its spec file directly under docs/app/screens/<persona>/<NN>-<slug>.md. The orchestrator then merges per-worker SCREEN_MAP rows and any new route declarations into the substrate index files. Expected wall-clock reduction vs sequential screen-spec: ~N/3 (Figma MCP calls dominate; partial parallelism in MCP throttling). Expected token reduction: ~2x (workers share no cross-screen context).

Mandatory context bootstrap (before any output):

<!-- shared:mandatory-context-bootstrap -->

• Read these sections in WORKFLOW_OPERATING_SYSTEM.md first:
  • ## LLM execution contract
  • ## Editor mode policy (mode definitions only; the tool mapping table is lazy-loaded in wos/editor-mode-mappings.md and needed only for non-Claude-Code tools)
  • ## Global output contract (including Adaptive handoff and Mode selection rule)
  • ## Cross-cutting workflow guardrails
• Bootstrap tiers (ADR-0025): the light-weight commands (branch-commit, what-next, where-we-at, slice-closure, compact-task-memory) may skip ## Editor mode policy good-fits lists and ## Cross-cutting workflow guardrails sequencing heuristics, reading only the mode definitions and the core guardrail rules (routing memory, command-less input triage, official command names, material change, no-op). The full tier is measured at 9610 tokens: the combined size of the four always-read WORKFLOW_OPERATING_SYSTEM.md sections listed above. The reduced tier is a self-declared estimate of about 3,500 tokens for the trimmed subset above; it has not been independently re-measured by the same method, and should be read as an estimate rather than a fresh figure. The same reduced tier extends to the high-frequency execution commands implement-approved-slice and sync-task-state (v3 wave1 item D: the most-invoked commands pay the bootstrap most often; state-reconcile deliberately stays on the full tier, cross-artifact judgment needs the full guardrail context).
• Cache-amortized layer (ADR-0006): this bootstrap floor is a cache-amortized cost, not a per-command tax paid in full on every invocation. It sits in the prompt cache for the session and is paid at write cost once per cache TTL window, then at roughly 0.1x on cached reads inside that window. Account for it separately from any per-skill Load budget (the generated .claude/skills/<name>/SKILL.md body); the two are different layers and should not be summed into one figure.
• Session bootstrap reuse (skip-if-unchanged; v3 wave1 item D): WHEN this same conversation already performed this bootstrap read in an earlier turn that is still VISIBLE in the current context window AND WORKFLOW_OPERATING_SYSTEM.md has not changed since, the command MAY skip the re-read and cite the earlier one instead, emitting one Command transcript line: Bootstrap: reusing turn <N> read, WOS unchanged. This is a scoped exception to the context-budget re-fetch rule (wos/context-budget.md, "The re-fetch rule"), justified because the bootstrap sections are one large, static, byte-identical read repeated every turn rather than a variable tool result; the re-fetch rule still governs every other tool result without exception. VISIBLE means the bootstrap section text itself is still present and quotable in the window right now, not merely that the record of an earlier read exists. On a harness that clears, a tool result can be emptied while the record that the tool ran survives (ADR-0114); a command that finds only that record, without the section text still readable, has not satisfied VISIBLE and must re-read. Self-declared memory after a compaction never qualifies (re-read instead), and a stateless-per-turn harness is excluded. The auditable-skip rule applies: the transcript line is mandatory; a silent skip is invalid output.
• Read additional sections only when relevant to this command's role.
• Read the commands/ directory command inventory to ensure command names and availability are current.
• Align all routing recommendations and next-command suggestions with the current command set.
• Official next-command names only: every recommended next command (including the handoff Run now line) MUST be the basename of an existing commands/<name>.md file in this workflow repository. Never invent names.

Required inputs:

• project workspace path
• Figma file URL (one per fleet run; all screens MUST belong to the same file)
• screens manifest: array of {screen_number, slug, persona, figma_node_id, journey?, route?} per screen (caller pre-enumerates; orchestrator does NOT auto-discover frames)
• persona scope: single persona for the whole fleet (mixing personas in one run is FORBIDDEN -- re-run per persona)
• path to SCREEN_MAP.md (default: docs/app/SCREEN_MAP.md; created from templates/SCREEN_MAP.md if absent)
• path to routes.md (default: docs/app/routes.md)
• optional: explicit max_fanout override (defaults to 20)

Task repository files to update:

• docs/app/screens/<persona>/<NN>-<slug>.md (one per worker; each worker is the SOLE writer of its file -- no overlap risk by construction since the orchestrator validates uniqueness of screen_number+persona+slug triples pre-dispatch)
• SCREEN_MAP.md (orchestrator is the SOLE writer per wos/substrate-peers.md; rows merged from worker partials)
• routes.md (orchestrator is the SOLE writer; rows appended for any new_route returned by workers)
• TASK_STATE.md ## Last completed step per the canonical 5-section write pattern in commands/_shared/task-state-slice-closure-pattern.md
• .wos/fleet-inbox/<run_id>/ directory (gitignored; one partial per worker)
• .wos/VERIFICATION_LOG.jsonl (one line per merged section + one per per-worker classification event)

Operating rules:

• Handoff: end with the adaptive ### Handoff block per WORKFLOW_OPERATING_SYSTEM.md ## Global output contract.
• Step 1: Validate manifest. Confirm N >= 6 (else NO_OP_TRACE: route to screen-spec sequential). Confirm one persona across all entries (else NO_OP_TRACE: list mixed personas; re-run per persona). Confirm every triple (screen_number, persona, slug) is unique (else NO_OP_TRACE: list duplicates). Confirm every figma_node_id is non-empty. Confirm N <= max_fanout (else NO_OP_TRACE: list overflow; suggest splitting by journey).
• Step 2: Verify prerequisites. SCREEN_MAP.md exists (or create from template); routes.md exists (or create with header); foundations dir present per wos/design-system-conventions.md; components dir present. If foundations are missing, NO_OP_TRACE: route to design-bootstrap.
• Step 3: Verify Figma MCP availability. Confirm get_design_context and get_screenshot tools are reachable; else NO_OP_TRACE: instruct caller to enable Figma MCP.
• Step 4: Verify tier guard. Orchestrator runs Sonnet (claude-sonnet-4-6); workers run Sonnet (claude-sonnet-4-6). Per wos/sub-agent-orchestration.md ## Tier-mapping per role: per-target deep analysis (one screen) -- Sonnet (orchestrator pays cost; correctness wins). Tier guard PASS (orch tier >= worker tier; equal is allowed).
• Step 5: Dispatch workers via the Workflow tool. For each manifest entry, dispatch a stateless sub-agent through the Workflow tool (the canonical parallel-orchestration primitive per ADR-0038). Pass task_input matching worker_input_schema. Each worker writes its spec file directly to <screens_root>/<persona>/<NN>-<slug>.md AND MUST invoke the StructuredOutput tool exactly once with artifact='<worker_id>.partial.json' and typed content matching worker_output_schema (per ADR-0038 Rule 1: structured output is mandatory; free-form prose is forbidden). The main loop persists the partial to active/<task>/.wos/fleet-inbox/<run_id>/<worker_id>.partial.json as typed JSON, not Markdown, so the apply step consumes typed data rather than re-parsing prose.
• Step 6: Each worker (instruction template). Worker executes the standard 12-step screen-spec flow (see commands/screen-spec.md Operating rules) for ONE screen: get_design_context -> get_screenshot -> identify components -> layout sketch -> spacing -> data -> copy -> a11y -> interactions -> error states -> related screens -> write file. Worker MUST NOT touch SCREEN_MAP.md or routes.md (substrate-peer rule); instead, return the screen_map_row + optional new_route via StructuredOutput. Final instruction to every worker: Return STRUCTURED OUTPUT with artifact='<worker_id>.partial.json' content=<worker_output_schema payload> where the payload is {status: "satisfied", spec_path: "<path>", screen_map_row: {...}, new_route: {...} | null, components_referenced: [...], components_candidate: [...], copy_strings: [...], open_questions: [...]}.
• Step 7: Wait for convergence. Barrier pattern: wait for all N workers OR timeout_ms (15 min default; Figma MCP latency can be high). Read all .partial.json files in active/<task>/.wos/fleet-inbox/<run_id>/. Classify per commands/_shared/convergence-policy.md failure table. Workers that did not invoke StructuredOutput (schema-skip) are classified failed and excluded from merge.
• Step 8: Merge SCREEN_MAP rows. Apply union merge: collect all screen_map_row entries from surviving typed partials. Deduplicate by (persona, screen_name) key. If a row already exists in SCREEN_MAP.md with the same key but different spec_doc or figma_node_id, REFUSE that row and log event=fleet-merge with conflict details. Sort merged output by persona ASC, then screen_number ASC. Per ADR-0038 Rule 2 (substrate writes sequenced through deterministic apply step in main loop), this merge runs serially in the orchestrator -- never in a worker.
• Step 9: Merge routes. Collect every non-null new_route from typed partials. Deduplicate by path key. If a route exists in routes.md with a different persona/screen, REFUSE that route and log conflict. Append surviving new routes to routes.md per its table convention.
• Step 10: Write SCREEN_MAP.md. Emit transaction header above the rows section; replace the section content per the canonical SCREEN_MAP table shape (Route | Persona | Screen name | Spec doc | Status | Figma frame | Notes). Preserve rows from outside this fleet's persona scope unchanged. Anchor every appended row to an existing parent heading or list per ADR-0038 Rule 3; refuse any row whose anchor is not located.
• Step 10.5: Scan substrate orphans. Run python3 scripts/scan-substrate-orphans.py docs/app/SCREEN_MAP.md docs/app/routes.md against the just-written files (ADR-0038 Rule 3: the apply step MUST detect and prevent the substrate-bullet-orphan failure mode per wos/bug-classes/substrate-bullet-orphan.md). On non-zero exit (one or more orphans detected): REFUSE the merge transaction, log event=fleet-merge-orphan-refused with the orphan list to VERIFICATION_LOG.jsonl, REVERT both SCREEN_MAP.md and routes.md to their pre-merge state, and surface the orphan list in ### Command transcript. On exit 0: proceed to Step 11.
• Step 11: Emit VERIFICATION_LOG.jsonl. One line per per-worker classification event (event=merge_include, event=worker_failed, event=worker_timeout, etc.; a worker that skipped for a missing Figma frame or schema is event=worker_missing with the skip cause in reason, NOT a bespoke schema_skip event, which is not in the canonical taxonomy) plus one line for the merged SCREEN_MAP section AND one for routes.md (event=fleet-merge, partials=[worker_id, ...], strategy=union).
• Step 12: Update TASK_STATE.md. Per the canonical 5-section write pattern. Include the fleet summary: total screens specced, persona, new routes appended, components referenced, components candidate (not yet in DS), top open questions.
• Workers NEVER write SCREEN_MAP.md or routes.md. Workers DO write their own spec file directly (no overlap by construction).
• Mixing personas in one fleet run is FORBIDDEN. Re-run per persona to keep merge keys unambiguous.
• Do NOT implement screens here. This command produces spec docs only; implementation flows through normal slice pipeline (task-init per screen group -> implementation-plan -> implement-approved-slice).

Required output:

1. Screen count + persona + journey scope
2. Dispatch summary: N dispatched, M satisfied, K needs_revision, L failed, P interrupted, T timed out, S skipped (no frame; logged worker_missing)
3. Merge summary: SCREEN_MAP rows merged + dedup count + conflict count; routes appended + conflict count; orphan-scan result (exit code + orphan list if any)
4. Components inventory: unique components referenced across all specs; candidates not yet in DS
5. Copy roll-up: total copy strings ready for i18n (count only; full list lives in spec files)
6. Top 5 open questions across the fleet
7. Path to updated SCREEN_MAP.md and list of new spec files
8. Recommended next command (typically journey-map if the fleet covered a complete journey, or task-init per screen group for implementation)

Claim grounding (active epistemic humility)

<!-- shared:claim-grounding -->

Claim grounding (active epistemic humility). This block governs what you may assert and how you record it. It is keyed to the substrate section you are writing, not to which command is running, and it is INERT on any output that writes none of the claim-bearing sections below. Full contract and rationale: wos/active-epistemic-humility.md.

1. When this applies. This block fires ONLY while you are writing a claim-bearing substrate section: TASK_STATE.md ## Current known facts, ## Risks to watch, ## Observations, ## Active files in scope, ## Canonical decisions; DECISIONS.md ## Locked decisions; IMPLEMENTATION_PLAN.md ## Current gaps, ## Risks and mitigations; IMPACT_ANALYSIS.md; EXTERNAL_RESEARCH.md; REFERENCES.md; or any section whose content is a statement a later command or a human decision will act on. WHEN your output writes none of these, this block imposes nothing: skip it and proceed. This is the D-13 inert clause; a fully-grounded or claim-free output pays nothing.

2. The unit is the load-bearing claim. A load-bearing claim is one a downstream command or a human decision consumes. A passing aside is not load-bearing; a statement someone will act on is. Apply the rest of this block per load-bearing claim, not per sentence.

3. Ground it or abstain. Before you assert a load-bearing claim, trace it to the enumerable grounded set: a captured REFERENCES.md entry, a file read in this session, command output actually seen, or a passing deterministic gate. A claim supported only by model memory is OUTSIDE the grounded set, including when you are right, because that support is not observable. WHEN a load-bearing claim falls outside the set, do NOT assert it: either investigate until it is grounded, or abstain per rule 6.

4. Status records provenance, never confidence. WHERE you attach an epistemic status to a claim, the status names WHERE THE CLAIM CAME FROM: a REFERENCES.md entry title, a file path plus line, or the gate output it came from. It SHALL NOT express a degree of certainty. Do NOT add a confidence field, a numeric threshold, or a self-assessment prompt anywhere; a self-reported confidence signal is not a usable control signal (wos/active-epistemic-humility.md Part 1.3). A status whose referent slot is empty is read as UNKNOWN, not as a weak yes.

5. Persisted claims carry the status; chat-only claims carry it when they route. Every load-bearing claim you write into a task-memory artifact carries its provenance referent, and that referent travels with the claim so a later command reads it too; do not drop it at the write boundary. A load-bearing claim that appears only in a chat-turn output carries a status only when it crosses the grounding boundary and triggers a route (an abstention, an escalation).

6. Abstain as a routed continuation, never a bare refusal. WHEN you abstain, name the specific investigation that would settle the question AND route to the command that runs it (capture-references, code-locate, incident-triage, or the fitting one). A withholding that stalls the work is invalid output. Abstention is distinct from NO_OP: NO_OP means there is no work to do; abstention means there is work and the grounding to do it is missing.

7. An unfired gate is not evidence. The absence of a fired check does not mean grounding existed. Do not read silence here as a pass.

Standard output layout (required)

<!-- shared:standard-output-layout -->

Produce the command output using this structure (English only):

Artifact changes

<!-- shared:artifact-changes-default -->

Follow ## Global output contract in WORKFLOW_OPERATING_SYSTEM.md for APPLIED / PROPOSED / SKIP rules.

Command transcript

<!-- shared:command-transcript-standard -->

Brief audit trail (max 4 lines; max 3 in no-op runs with NO_OP_TRACE).

Handoff

<!-- shared:handoff-body -->

Use the adaptive ending format from WORKFLOW_OPERATING_SYSTEM.md ## Global output contract (Mode A compact or Mode B full per session state).

Definition of done (command output)

• Every manifest entry has exactly one spec file written under docs/app/screens/<persona>/<NN>-<slug>.md.
• SCREEN_MAP.md has one row per surviving worker output; conflicts REFUSED and logged.
• routes.md has new routes appended (one per new_route returned); conflicts REFUSED and logged.
• All 12 sections of the SCREEN_SPEC template are filled per spec file.
• Per-worker partials persisted as typed JSON in .wos/fleet-inbox/<run_id>/<worker_id>.partial.json (never .partial.md).
• Every worker invoked the StructuredOutput tool exactly once with the declared schema; schema-skip workers (no Figma frame) logged as event=worker_missing with the skip cause in reason and excluded from merge (ADR-0038 Rule 1).
• scripts/scan-substrate-orphans.py returned exit 0 against SCREEN_MAP.md and routes.md after the merge (ADR-0038 Rule 3). Non-zero exit triggers REVERT and event=fleet-merge-orphan-refused.
• VERIFICATION_LOG.jsonl has one line per classification event + one per merged section + one for the orphan-scan result.
• Worker contract violations (mid-flight writes to SCREEN_MAP / routes) explicitly listed in ### Command transcript.
• No code implemented by this command; output explicitly says "produces specs + index only; implementation flows through task-init".
• Substrate peer rule respected: SCREEN_MAP.md and routes.md are owned by this command in fleet mode (single-screen screen-spec retains co-ownership for sequential runs; mixed-mode is reconciled by state-reconcile).
• Before declaring this output done, confirm it satisfies the shared Definition of done (command outputs) and Gate conditions in WORKFLOW_OPERATING_SYSTEM.md.

Quality bar: This is the J.7 PILOT -- the second real orchestrator under ADR-0034, and the maintainer's primary multi-agent scenario (designer hands off a 12-30 screen Figma file; specs land in minutes not hours). A developer reading any merged spec doc plus the components it references must be able to build the screen without opening Figma. Silent dropping is forbidden: every manifest entry either produces a spec file with all 12 sections OR appears explicitly in the failure classification with a reason. The fleet is the eval-baseline for K.7 -- the harness compares wall-clock + tokens + per-spec completeness against screen-spec sequential on identical manifests. This command MUST comply with ADR-0038 (docs/adr/0038-workflow-tool-as-parallel-orchestration-primitive.md): Workflow tool as the dispatch primitive (Rule 1: structured output mandatory via StructuredOutput tool), parallel-then-sequential-apply (Rule 2: substrate writes sequenced through the orchestrator main loop), and post-apply orphan scan via scripts/scan-substrate-orphans.py (Rule 3: prevent the wos/bug-classes/substrate-bullet-orphan.md failure mode that produced 8 orphans in pilot-repo during K.2).

<!-- cache-breakpoint -->