Atom audit fleet

Act as a senior/staff design system orchestrator dispatching N atom-auditor sub-agents and synthesizing their partials into the canonical ATOM_AUDIT.md table.

Goal: Audit every atom in packages/design-system/src/atoms/ against docs/research/COMPONENT_GUIDELINES.md in parallel: dispatch N Haiku workers (3-5 atoms per worker), wait for convergence, merge their structured rows into ATOM_AUDIT.md. ~10x token reduction vs atom-audit single-agent expected when atom count >= 6, because Haiku per-token cost is much lower and the rule checks are mechanically schema-bounded.

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
• path to atoms directory (default: packages/design-system/src/atoms/)
• path to COMPONENT_GUIDELINES.md (default: docs/research/COMPONENT_GUIDELINES.md)
• path to ATOM_AUDIT.md (default: docs/research/ATOM_AUDIT.md; created from templates/ATOM_AUDIT.md if absent)
• optional: explicit max_fanout override (defaults to 20)
• optional: explicit per-worker batch size (default 4; range 3-5)

Task repository files to update:

• ATOM_AUDIT.md (sole owner of merged result; the orchestrator is the SOLE writer per wos/substrate-peers.md)
• 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: Enumerate atoms. List every directory under the atoms path. For each, locate the main component file (<Name>/index.tsx or <Name>/<Name>.tsx). Filter out non-atom artifacts (test files, story files, type-only files).
• Step 2: Compute batches. Partition atom paths into batches of batch_size (default 4). N batches = N workers. If N == 0, NO_OP_TRACE: nothing to audit. If N > max_fanout, STOP with NO_OP_TRACE listing the overflow; recommend running on a subdirectory first.
• Step 3: Verify prerequisites. Confirm COMPONENT_GUIDELINES.md exists; ATOM_AUDIT.md exists (or create from template). If COMPONENT_GUIDELINES.md is missing, NO_OP_TRACE and route to design-bootstrap.
• Step 4: Verify tier guard. Orchestrator runs Sonnet-class; workers run Haiku-class (both per the suggested-model frontmatter, not pinned in prose, so a model-generation bump updates one field instead of the body). Orchestrator tier >= worker tier per wos/sub-agent-orchestration.md ## Tier-aware dispatch protocol. PASS.
• Step 5: Dispatch workers. For each batch, invoke a stateless sub-agent via the host's primitive (Claude Code Task tool with subagent_type: general-purpose and a Haiku-class tier hint per suggested-model). Pass task_input matching worker_input_schema: {atom_paths: [...], guidelines_path: "<path>"}. Each worker MUST return its result via the StructuredOutput tool keyed artifact=fleet-inbox/<run_id>/<worker_id> (ADR-0038 Rule 1; prose .partial.md returns FORBIDDEN, a typed .partial.json is replay-only) per the worker contract.
• Step 6: Each worker (instruction template). Worker reads guidelines_path; for each atom in atom_paths, reads the main component file; mechanically checks: memo (is React.memo / forwardRef-memo wrap present? prop count threshold), callbacks (count of inline arrow callbacks not wrapped in useCallback), inline_styles (count of object-literal style={{...}}), press_anim (useAnimatedPress vs useState transform if press handler present), touch_target (44pt iOS / 48dp Android minimum if interactive), a11y (accessibilityRole + accessibilityLabel for icon-only buttons + accessibilityState for interactive variants), reduced_motion (useReducedMotion() check if transform/translate animation). Sum failing rules into changes_needed. Return {status: "satisfied", rows: [...]}.
• Step 7: Wait for convergence. Barrier pattern: wait for all N workers to terminate OR timeout_ms (10 min default) to elapse. Read all files in active/<task>/.wos/fleet-inbox/<run_id>/. Classify per commands/_shared/convergence-policy.md failure table.
• Step 8: Merge. Apply union merge strategy: collect all rows from all surviving partials; deduplicate by component key (each atom audited by exactly one worker; duplicates would indicate a bug -- log event=fleet-merge warning with partials=[...]). Sort rows by changes_needed descending then component ascending (highest-impact fixes surface first).
• Step 9: Write ATOM_AUDIT.md. Emit transaction header above the table section; replace the ## Summary Table section content with the merged rows; append a new row to ## Audit history with date + total changes_needed sum + cleared delta vs previous run.
• Step 10: Emit VERIFICATION_LOG.jsonl. One line per per-worker classification event (event=merge_include, event=worker_failed, event=worker_timeout, etc.) plus one line for the merged section (event=fleet-merge, partials=[worker_id, ...], strategy=union).
• Step 10.5: Scan substrate orphans (ADR-0038 Rule 3 gate). After the substrate write in Step 9 and the VERIFICATION_LOG emission in Step 10, invoke python3 scripts/scan-substrate-orphans.py <ATOM_AUDIT.md path> <TASK_STATE.md path> against every file this command touched. On non-zero exit code: roll back the ## Summary Table section replacement in ATOM_AUDIT.md (restore the pre-write snapshot), append a line event=orphan_detected (with files=[...] and exit_code=<n>) to .wos/VERIFICATION_LOG.jsonl, and return NO_OP_TRACE routing to manual repair per wos/bug-classes/substrate-bullet-orphan.md. On exit code 0, proceed to Step 11. The orphan-scan gate is non-negotiable per ADR-0038 Rule 3.
• Step 11: Update TASK_STATE.md. Per the canonical 5-section write pattern. Include the audit summary: total atoms audited, total changes_needed, top-3 fix groupings (rules with most failing atoms).
• Workers NEVER write substrate, because parallel workers writing the same file would race and corrupt the merged result and scramble provenance; routing every write through the orchestrator's one apply step keeps the merge deterministic and attributable (ADR-0038 Rule 2). The orchestrator is the SOLE writer of ATOM_AUDIT.md.
• Do NOT implement fixes here. This command produces the audit only; fixes flow through normal slice pipeline (task-init per fix grouping -> impact-analysis -> implementation-plan -> implement-approved-slice).
• If COMPONENT_GUIDELINES.md added a new rule not represented in the worker_output_schema columns, NO_OP_TRACE: route to a schema-extension slice first (worker_output_schema update + ATOM_AUDIT.md column add).

Required output:

1. Atom count enumerated + batch breakdown (N workers, atoms per worker)
2. Dispatch summary: N dispatched, M satisfied, K needs_revision, L failed, P interrupted, T timed out
3. Merge summary: total rows merged, dedup count, conflict count
4. Per-rule failure breakdown: which rules have most failing atoms
5. Top 3 suggested fix groupings (each a candidate slice for task-init)
6. Path to the updated ATOM_AUDIT.md
7. Recommended next command (typically task-init for the highest-priority fix grouping)

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 atom under the atoms path has exactly one row in the merged ATOM_AUDIT.md table.
• Every rule from COMPONENT_GUIDELINES.md is represented as a column; new rules trigger NO_OP_TRACE schema-extension routing.
• changes_needed integer per row matches the count of failing rules in that row.
• Audit history row appended for this run with date + total + cleared delta.
• Top 3 fix groupings suggested in the output (NOT in the file).
• Worker contract violations explicitly listed in ### Command transcript.
• No code fixes applied by this command; output explicitly says "produces audit only, fixes flow through task-init".
• Substrate peer rule respected: ATOM_AUDIT.md ## Summary Table and ## Audit history are owned by this command per the substrate ownership matrix (folded entry added in K.2 retrofit; until then, this command is the de-facto owner via templates/ATOM_AUDIT.md convention).
• scan-substrate-orphans.py exit code 0 on every touched file (ATOM_AUDIT.md, TASK_STATE.md). A non-zero exit blocks completion, triggers Step 10.5 rollback, and routes to manual repair per wos/bug-classes/substrate-bullet-orphan.md.
• 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.6 PILOT -- the first real orchestrator under ADR-0034, now also bound by ADR-0038 (Workflow tool as canonical parallel-orchestration primitive). The three ADR-0038 rules apply: (1) structured output is mandatory (enforced via worker_output_schema), (2) substrate writes sequence through a deterministic apply step (Step 8 merge + Step 9 sole-writer replace), (3) the apply step MUST detect and prevent the substrate-bullet-orphan failure class per wos/bug-classes/substrate-bullet-orphan.md -- gated by Step 10.5 invoking scripts/scan-substrate-orphans.py. If the orchestrator cannot determine which row belongs to which atom (duplicate component keys across partials, or atom not represented), it refuses to write that row and flags the ambiguity in ### Command transcript. Silent dropping is forbidden. The pilot is the eval-baseline for K.7 -- the harness compares this command's output against atom-audit single-agent on identical inputs to validate the ~10x cost-reduction claim.

<!-- cache-breakpoint -->