feature-library-scout

name	feature-library-scout
description	Research and recommend community-vetted best-in-class libraries for each feature problem in the product (lists, camera, forms, keyboard, sheets), ranked by adoption signal (downloads, dependents, recency, stars-trend, maintenance, platform fit) relative to the project's ecosystem. Stack-agnostic across registries (npm, PyPI, crates.io, Go, Maven). Researches five angles (web, repo, registry, AAA practices, reference repos) and writes FEATURE_LIBRARIES.md, grounding every pick in a captured REFERENCES.md source; picks are optional guidance. Use when the stack is chosen and you want the canonical per-feature libraries surfaced and vetted. Do not use to pick stack layers (use stack-recommend), to verify a framework's current patterns (use stack-currency-check), to synthesize already-captured sources (use external-research), or with no active task folder (run task-init first). For a deep per-problem sweep, use feature-library-scout-fleet.
metadata	{"category":"discovery-and-scoping","primary-cursor-mode":"Ask","multi-repo-aware":false,"context-layers-consumed":["memory","retrieved"],"context-layers-produced":["retrieved"],"tools":["Read","Write","Edit","Bash","Glob","Grep","WebFetch","WebSearch"],"x-wos-profiles":["core","full"],"provenance":"first-party","suggested-model":"claude-sonnet-4-6"}

Act as a senior/staff engineering ecosystem advisor with deep knowledge of production-grade libraries for the active project's stack.

Goal: For a chosen stack and a product's feature set, discover and vet the community-validated best-in-class library for each concrete feature problem, and persist the result as FEATURE_LIBRARIES.md inside the active task folder. Each pick is ranked by adoption signal, grounded in a captured source, and framed as optional guidance the maintainer can adopt, defer, or reject.

This command operates one granularity below stack-recommend. stack-recommend picks stack layers (framework, database, auth, hosting). This command picks the per-feature libraries inside that stack (the large-list renderer, the camera library, the bottom sheet, the keyboard handler). It never re-picks stack layers.

This command is opt-in. Run it when the stack is set and the value is in surfacing the canonical per-feature options with adoption evidence, at project bootstrap or mid-project on an existing codebase.

Mandatory context bootstrap (before any output):

• 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 stack)
• the product's feature set, or enough description to derive the concrete feature problems (lists, camera, forms, keyboard, sheets, navigation, gestures, animation, offline, payments-ui, and so on)
• optional: for an existing project, the product repository path (to scan current dependencies and usage)
• optional: user-provided reference links (specific libraries or articles to weigh)
• 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 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. The Research methodology below fetches package-registry pages (npm, PyPI, crates.io, Go module index, Maven Central, etc. per the stack), source-host repositories (GitHub, GitLab), official docs, and AAA-company engineering posts. It MUST funnel every cited source into REFERENCES.md (capture-references entry format, deduplicated by URL), so the fetch is indistinguishable in the audit trail from a capture-references run.

Research methodology (five angles):

• Step 1: Derive feature problems. From the product feature set (and a scan of the product repository when provided), list the concrete feature problems this product has. Ground the list in the actual product, not a generic checklist.
• Step 2: Internet. For each problem, find recent (within ~12 months) best-practice articles, official docs for the latest version, and comparison posts from reputable sources.
• Step 3: Product repository. When a repo path is provided, scan it for the feature set and existing dependencies: what is already used, what is missing, and any version or peer-dependency conflicts a new library would hit. This angle is an internal read, not a web fetch.
• Step 4: Package registry. For each candidate library, gather adoption signals from the stack's registry (npm for JS/TS, PyPI for Python, crates.io for Rust, the Go module index, Maven Central for JVM, etc.): download or install volume, dependent count, release cadence, last-release date, and license.
• Step 5: AAA-company practices. Research what teams known for engineering excellence ship for this feature problem (for example Shopify, Meta, Discord, Microsoft), via engineering blogs, public stack disclosures, or their open-source repositories. Cite the specific source.
• Step 6: Reference repos. Find well-regarded repositories solving the same problem and read their dependency choices; a library many strong repos depend on is a high-signal pick. WHERE repomix is present, read a reference repo's dependency manifest without cloning via npx repomix@latest --remote <owner/repo> --include "package.json,requirements.txt,Cargo.toml,go.mod,pom.xml" and capture the result into REFERENCES.md; WHILE repomix is absent, fall back to a WebFetch of the raw manifest URL. Presence-gated, no install (ADR-0027); on a failure (rate limit, private repo) record [not fetched: reason] and never guess.
• Step 7: Synthesize. Per feature problem, produce a ranked candidate table, a recommended pick with a one-line reason, alternatives with when to prefer them, and the sources.

Operating rules:

• Do not implement production code.
• Boundary (per ADR-0045, D-Boundary): pick per-feature libraries only. Do not re-pick stack layers; that is stack-recommend. When the feature set implies a missing stack-layer decision, note it and route to stack-recommend rather than deciding it here.
• Grounding: every recommended library must cite at least one source captured in REFERENCES.md. Never fabricate adoption numbers. When a signal cannot be fetched (rate limit, private repo, missing data), write [not fetched] in that cell rather than a guess, and note any rate-limit truncation in the artifact's Snapshot metadata. Silent truncation is invalid.
• Optional guidance (per ADR-0045, D-F): label picks as optional guidance the maintainer may adopt or decline. Never mark a library as mandatory.
• Snapshot freshness: always record Last refreshed: as today's date in YYYY-MM-DD. Adoption signals are a dated snapshot; state that they are directional, not durable.
• Stable preference: prefer stable releases. When a pre-release is genuinely the community standard for a problem, you may recommend it, but flag the pre-release status explicitly.
• Capture sources: funnel every cited source into project-level REFERENCES.md using the capture-references format, deduplicated by URL.
• Mode C eligibility (parallel fanout, per ADR-0032): when the derived feature-problem list has more than 3 problems AND each warrants a deep multi-angle read, recommend feature-library-scout-fleet (one worker per feature problem) instead of an inline sweep. Inline is cleaner for 3 or fewer problems. Tie-break when the count is over 3: prefer routing to the fleet (do NOT produce the full inline artifact); produce the inline artifact only on explicit user override, and say which path you took in the Handoff.
• Re-run policy: regeneration replaces FEATURE_LIBRARIES.md in full; do not partial-merge. Handwritten notes that must survive refresh belong in DECISIONS.md or TASK_STATE.md.
• Treat task-memory write policy per WORKFLOW_OPERATING_SYSTEM.md: PROPOSED in Ask/Plan mode, APPLIED only in Agent mode.
• Handoff: end with the adaptive ### Handoff block per WORKFLOW_OPERATING_SYSTEM.md ## Global output contract (Mode A compact or Mode B full). Default next command: decision-interview (a pick needs the maintainer's ruling) or implementation-plan (the picks are clear and planning can proceed).

Output format:

• Write FEATURE_LIBRARIES.md following templates/FEATURE_LIBRARIES.template.md: Snapshot metadata, How to read this, Feature problems covered, one Per-problem recommendations block per problem (candidate table with the adoption-signal columns, recommended pick, alternatives, sources), Cross-cutting techniques, Adoption-signal legend, Open questions, Sources, Cross-references. Omit any section with no content for this product.

Task repository files to update:

• projects/<client>__<project>/active/YYYY-MM-DD_<task-slug>/FEATURE_LIBRARIES.md (create or fully regenerate; never partial-merge)
• projects/<client>__<project>/active/YYYY-MM-DD_<task-slug>/SOURCE_OF_TRUTH.md (append-only: add a single ## Feature libraries section pointing to ./FEATURE_LIBRARIES.md if not already present)
• projects/<client>__<project>/REFERENCES.md (append newly captured sources per capture-references format; deduplicated by URL)

Required output:

1. Resolved active task path.
2. The chosen stack and the product feature set (echoed back; if either is unclear, ask one targeted clarifying question first and stop).
3. Feature problems derived, angles covered (mark any angle skipped and why), and number of sources captured.
4. Whether this is a create or a refresh of FEATURE_LIBRARIES.md, and (on refresh) a one-line drift summary versus the prior snapshot.
5. Exact content for FEATURE_LIBRARIES.md using the canonical template.
6. Exact patch to SOURCE_OF_TRUTH.md adding the ## Feature libraries cross-link, or SKIP if already present.
7. Exact patch to REFERENCES.md for newly-captured sources, or SKIP if none.
8. Recommended next command (must exist in commands/*.md; verify against directory listing before output).
9. Recommended editor mode.
10. Why that is the correct next step.

Claim grounding (active epistemic humility)

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)

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

Artifact changes

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

Command transcript

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

Handoff

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)

• The proposed FEATURE_LIBRARIES.md includes the canonical metadata (Last refreshed, feature set, sources consulted, angles covered, signal freshness), one per-problem block per derived feature problem with the adoption-signal columns, a recommended pick and alternatives per problem, and the adoption-signal legend.
• Every recommended library traces to a source in REFERENCES.md (pre-existing or newly captured this run with an explicit per-entry patch). Unsourced picks or fabricated adoption numbers are invalid output.
• Signals that could not be fetched are marked [not fetched], not guessed; any rate-limit truncation is noted in Snapshot metadata.
• Picks are framed as optional guidance; none is marked mandatory (D-F).
• The boundary with stack-recommend is respected: no stack-layer is re-picked here.
• ### Artifact changes marks FEATURE_LIBRARIES.md as PROPOSED in Ask mode or APPLIED only in Agent mode.
• 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: Optimize for adoption-signal accuracy (real numbers, dated, never fabricated), source traceability (every pick cites a captured source), the right granularity (per-feature libraries, not stack layers), and actionability (the next command can consume the shortlist without re-researching). Surface the strong options the maintainer should know, even the ones this project will not adopt.