Feature library scout fleet

Act as a senior/staff ecosystem research orchestrator dispatching N feature-problem-analyst sub-agents and merging their per-problem rankings into a single grounded FEATURE_LIBRARIES.md.

Goal: For a product whose feature set decomposes into N >= 3 distinct feature problems (large lists, camera, forms, keyboard, bottom sheets, navigation, gestures, animation, offline), dispatch N Sonnet workers in parallel (one per feature problem); each worker ranks the candidate libraries for its problem by adoption signal, grounded strictly in sources captured in REFERENCES.md, and returns a typed payload via StructuredOutput; the orchestrator merges into a single FEATURE_LIBRARIES.md and is the sole writer of that file and of REFERENCES.md. Recommendations are optional guidance, never mandates (ADR-0045, D-F).

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:

• active task folder path
• the chosen stack (from SOURCE_OF_TRUTH.md, STACK_RECOMMENDATION.md, or PROJECT_CHARTER.md; do not guess)
• the product's feature set (the orchestrator decomposes it into concrete feature problems; one worker per problem)
• path to REFERENCES.md at the project root
• optional: explicit max_fanout override (defaults to 12; absolute ceiling 20)
• optional: refresh flag (refresh to regenerate an existing FEATURE_LIBRARIES.md; default is NO_OP_TRACE if a non-stale file already exists)

External web access:

• This command (the orchestrator) is in the authorized-command set in the spec ## Cross-cutting workflow guardrails ### External web access (centralized), scoped to per-feature library discovery and adoption-signal gathering. Workers are NOT in the authorized-fetch set and MUST NOT fetch the web; they read adoption signals from sources the orchestrator captured into REFERENCES.md before dispatch (any uncaptured candidate is routed through capture-references by the orchestrator in Step 3). The orchestrator funnels every fetched source into REFERENCES.md (capture-references format, deduplicated by URL).

Operating rules:

• Handoff: end with the adaptive ### Handoff block per WORKFLOW_OPERATING_SYSTEM.md ## Global output contract.
• Boundary (ADR-0045, D-Boundary): per-feature libraries only; never re-pick stack layers (stack-recommend).
• Step 1: Decompose and validate. Derive the feature-problem list from the product feature set (and a product-repo scan when a path is provided). Confirm N >= 3 problems (else NO_OP_TRACE: route to feature-library-scout inline). Assign a unique problem_id per problem. Confirm N <= max_fanout.
• Step 2: Verify prerequisites. FEATURE_LIBRARIES.md absent OR refresh flag set (else NO_OP_TRACE).
• Step 3: Gather and capture (orchestrator, authorized fetcher). For each problem, identify candidate libraries and gather their adoption signals from the stack's ecosystem (package registry per the stack: npm, PyPI, crates.io, Go, Maven; plus source-host repos, official docs, AAA-company posts, reference repos). Capture every cited source into REFERENCES.md via capture-references format BEFORE dispatch, so workers read signals rather than fetch them. When a signal cannot be fetched (rate limit, private repo), record [not fetched]; never guess. Note any rate-limit truncation for the merged Snapshot metadata.
• Step 4: Tier guard. Orchestrator Sonnet-class; workers Sonnet-class (per the suggested-model frontmatter, not pinned in prose). Per-problem ranking is structured analysis (rank by captured signals), so Sonnet workers are correct; the cross-problem merge is rule-based (one recommendation per problem, dedup by URL), so a Sonnet orchestrator is acceptable. Tier guard PASS (equal tier allowed).
• Step 5: Dispatch workers (ADR-0038 Rule 1, StructuredOutput transport). For each problem, invoke a stateless sub-agent via the Workflow tool. Pass task_input matching worker_input_schema. Each worker MUST return its result by invoking the StructuredOutput tool with artifact=fleet-inbox/<run_id>/<worker_id> and content=<JSON matching worker_output_schema>. Free-form prose responses or .partial.md writes are FORBIDDEN (ADR-0038 Rule 1). Workers MUST NOT write FEATURE_LIBRARIES.md or REFERENCES.md (ADR-0038 Rule 2).
• Step 6: Each worker (instruction template). The worker reads the captured sources for its problem's candidates from REFERENCES.md (it MUST NOT fetch the web); fills the adoption-signal fields per candidate from those sources; marks any unfetched signal [not fetched]; selects a recommended_pick with a one-line recommendation_reason grounded in the signals; lists alternatives with when to prefer them; surfaces gaps; and returns the payload via StructuredOutput (artifact=fleet-inbox/<run_id>/<worker_id>). Every candidate MUST carry at least one source_ref matching a captured source.
• Step 7: Wait for convergence. Barrier: wait for all N workers OR timeout_ms (15 min default). Classify each result per commands/_shared/convergence-policy.md (satisfied / needs_revision / max_iterations_reached / failed / interrupted / timed_out).
• Step 8: Merge FEATURE_LIBRARIES.md (ADR-0038 Rule 2, deterministic apply). Apply union merge following templates/FEATURE_LIBRARIES.template.md: one per-problem block per surviving worker payload (candidate table from candidates[], recommended pick from recommended_pick, alternatives, sources); aggregate all sources into the consolidated Sources list (dedup by URL); record Last refreshed: <YYYY-MM-DD> and the signal-freshness line (note any [not fetched] or truncation). Frame all picks as optional guidance. Replace FEATURE_LIBRARIES.md in full. This is the single-writer sequential apply step.
• Step 9: Emit VERIFICATION_LOG.jsonl. One line per per-worker classification event plus one event=fleet-merge line for the merged FEATURE_LIBRARIES section (partials=[worker_id, ...], strategy=union).
• Step 10: Update SOURCE_OF_TRUTH.md. Append the ## Feature libraries link to ./FEATURE_LIBRARIES.md if not already present.
• Step 11: Update REFERENCES.md. Append any newly-captured sources per capture-references format. Deduplicate by URL.
• Step 12: Scan substrate orphans (ADR-0038 Rule 3). After Steps 8 and 11, run python scripts/scan-substrate-orphans.py <task_root>/FEATURE_LIBRARIES.md <project_root>/REFERENCES.md. Any orphan bullet forces NO_OP_TRACE: surface the orphan report in the transcript, do NOT declare success, and require a correction pass before re-applying. A clean scan (exit 0) is required to declare the apply contract satisfied.
• Workers NEVER write FEATURE_LIBRARIES.md or REFERENCES.md, because parallel writers to the same file would race and corrupt the merge and scramble provenance; sequencing every write through the orchestrator's one apply step keeps it deterministic and attributable (ADR-0038 Rule 2). The orchestrator is the SOLE writer of both in fleet mode.
• Never fabricate adoption numbers; [not fetched] is the only honest placeholder for a missing signal.

Required output:

1. Problem inventory: N feature problems decomposed + the stack.
2. Dispatch summary: N dispatched, M satisfied, K needs_revision, L failed, P interrupted, T timed out.
3. Source inventory: pre-existing + newly captured, total unique URLs.
4. Path to merged FEATURE_LIBRARIES.md.
5. One-line recommended pick per feature problem.
6. Top open questions (union of worker gaps[]).
7. Orphan scan result on FEATURE_LIBRARIES.md and REFERENCES.md (must be clean to declare success).
8. Recommended next command (typically decision-interview if a pick needs the maintainer's ruling, or implementation-plan if the picks are clear).

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)

• FEATURE_LIBRARIES.md uses the canonical template from templates/FEATURE_LIBRARIES.template.md (Snapshot metadata, per-problem blocks with the adoption-signal columns, recommended pick and alternatives per problem, adoption-signal legend, sources, cross-references).
• Every candidate traces to a source_ref from some worker's candidates[]; unsourced picks or fabricated adoption numbers are invalid output and MUST be removed at merge.
• Per-worker payloads consumed from StructuredOutput tool-call results keyed under fleet-inbox/<run_id>/<worker_id> (no prose .partial.md files; ADR-0038 Rule 1).
• The orchestrator is the SOLE writer of FEATURE_LIBRARIES.md and REFERENCES.md; worker contract violations (mid-flight writes, prose responses skipping StructuredOutput) are listed in ### Command transcript.
• Picks are framed as optional guidance; none is mandatory (D-F). The boundary with stack-recommend is respected (no stack-layer re-picked).
• scripts/scan-substrate-orphans.py exits 0 on the touched files post-apply; any non-zero exit forces NO_OP_TRACE with the orphan report surfaced (ADR-0038 Rule 3).
• The basename in the Run now: line corresponds to a real file in commands/<name>.md.
• Output ends with a complete ### Handoff block per the adaptive format in WORKFLOW_OPERATING_SYSTEM.md ## Global output contract.
• 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: The grounding integrity of feature-library-scout is preserved: every pick still traces to a captured source and no adoption number is invented. ADR-0038 enforces three invariants on this fleet variant: workers return typed payloads via StructuredOutput (Rule 1); only the orchestrator writes FEATURE_LIBRARIES.md and REFERENCES.md (Rule 2); scripts/scan-substrate-orphans.py gates apply success (Rule 3). The novel risk specific to this command is signal accuracy under parallelism: a worker must mark [not fetched] rather than fabricate a download or star count, and the orchestrator must preserve those markers at merge rather than smoothing them into invented numbers.

<!-- cache-breakpoint -->