docs(repo): align documentation scopes and skill tooling

.agents/skills/luwu-boundary-review/SKILL.md

@@ -44,3 +44,32 @@ concrete path/line and test evidence, failure impact, and minimum fix. Classify
 implemented, partial, design-only, or unstarted behavior and require a focused
 regression test for every changed shield. Use `docs/reference.md` for the
 stable write contract and `docs/design.md` for its mechanism.
+
+## Scope boundary
+
+This stance owns filesystem and resource-boundary questions: path roots,
+symlinks, descriptors, permissions, replacement, cleanup, and recoverability.
+It does not own who is authorized to change a value, whether an action had
+consent, what a representation means, or whether a claim is current. Hand
+those findings to the ownership, consent, semantic-drift, or verifiability
+stance instead of redefining their contracts here.
+
+## Good patterns
+
+- Re-observe staged no-follow identity at a replace boundary and classify an
+  uncertain result as unknown instead of inferring it from equal bytes.
+  Evidence: [M3 repair record](../../../docs/milestones/m3-repair-plan.md).
+- Check journal and lock paths together before either persistent entry is
+  created. Evidence: [M3 follow-up record](../../../docs/milestones/m3-followup.md).
+
+## Bad patterns
+
+- Treat equal target content as proof that this process performed the
+  replacement. This loses provenance at the exact boundary where the writer
+  may have raised.
+- Create a journal lock or temporary entry before checking its full path
+  relationship with declared resources. This can mutate a declared target
+  before the requested operation begins.
+
+Patterns are review evidence, not a replacement for the public write contract.
+Promote a repeated pattern only in the owning document and link back here.

.agents/skills/luwu-confidentiality-review/SKILL.md

@@ -46,3 +46,31 @@ Report the value flow, exposure boundary, path/line and test evidence, rejected
 diagnostic convenience, minimum fix, and regression test. Classify implemented,
 partial, design-only, or unstarted. Use the owning reference and provider
 contract for exact secrecy promises.
+
+## Scope boundary
+
+This stance owns sensitive-value and capability exposure: secrets, rendered
+content, metadata, logs, subprocesses, networks, caches, backups, and error
+surfaces. It does not own filesystem replacement correctness, authorization
+transitions, configuration authority, semantic equivalence, or documentation
+status. Refer those questions to the boundary, consent, ownership,
+semantic-drift, or verifiability stance.
+
+## Good patterns
+
+- Require explicit runtime authority, a bounded provider adapter, and fixed
+  safe error projections before a provider can run. Evidence: the M4 closure
+  record and the v6 contract in `docs/reference.md`.
+- Keep journals and caches metadata-only, and use a fake provider or injected
+  resolver for tests rather than a real vault. Evidence:
+  [M4 closure record](../../../docs/milestones/m4.md).
+
+## Bad patterns
+
+- Persist provider references, rendered bytes, content hashes, or secret
+  derived metadata in a journal or cache merely to improve diagnostics.
+- Run a real vault in a regression test or expose provider stdout/stderr in
+  an error, diff, log, or machine-readable result.
+
+Patterns are evidence for this review stance. The exact secrecy and capability
+contract remains owned by `docs/reference.md`.

.agents/skills/luwu-consent-review/SKILL.md

@@ -44,3 +44,30 @@ Report a short state-transition timeline with path/line and test evidence,
 classify behavior as implemented, partial, design-only, or unstarted, and name
 the smallest corrective change plus its regression test. Use the owning CLI or
 reference contract for exact exit codes and output promises.
+
+## Scope boundary
+
+This stance owns authorization transitions: observation, planning,
+confirmation, stale revalidation, mutation, partial completion, and the next
+safe action. It does not own path safety, secret exposure, configuration
+authority, semantic equivalence, or document truth. Hand those findings to the
+boundary, confidentiality, ownership, semantic-drift, or verifiability stance.
+
+## Good patterns
+
+- Keep inspect and plan read-only, require explicit confirmation for mutation,
+  and recalculate after the write. Evidence: the command contract in
+  `docs/reference.md`.
+- Complete preflight before the first writer and keep recovery observation
+  read-only. Evidence: the M3 execution closure and v5 reference contract.
+
+## Bad patterns
+
+- Let inspection, planning, recovery, or cache status perform an implicit
+  write or refresh.
+- Allow a selected reverse-sync field to proceed while another field in the
+  same resource requires conflict review. Evidence:
+  [M3 execution closure](../../../docs/milestones/m3-execution-closure.md).
+
+Patterns do not define confirmation flags or exit codes. Those remain owned by
+the public CLI and reference contract.

.agents/skills/luwu-ownership-review/SKILL.md

@@ -45,3 +45,32 @@ the protected authority, the convenience being sacrificed, path/line evidence,
 the failure scenario, the smallest corrective change, and the regression test.
 Use `docs/product.md` for the responsibility questions and the owning contract
 for the exact transition rules.
+
+## Scope boundary
+
+This stance owns responsibility and authority: declared owner, scope,
+provenance, adoption, baseline role, provider/live role, and reverse-sync
+eligibility. It does not own the filesystem mechanics, authorization prompt,
+secret exposure, semantic equivalence, or current-document status. Route those
+findings to the boundary, consent, confidentiality, semantic-drift, or
+verifiability stance.
+
+## Good patterns
+
+- Treat an absent baseline as `unbased` and a change outside declared fields
+  as a separate undeclared signal; neither silently grants a candidate.
+  Evidence: the M3 contract in `docs/reference.md`.
+- Permit reverse sync only through an explicit declared literal mapping and
+  preserve unselected or undeclared source content. Evidence:
+  [M3 repair record](../../../docs/milestones/m3-repair-plan.md).
+
+## Bad patterns
+
+- Use a generic force or precedence rule to adopt undeclared live content or
+  to erase a conflict.
+- Re-read a convenient baseline and use it as authorization when the planner
+  classified a different baseline. Evidence:
+  [M3 follow-up record](../../../docs/milestones/m3-followup.md).
+
+Patterns explain how to attack authority mistakes; the resource and field
+contract remains owned by `docs/product.md` and `docs/reference.md`.

.agents/skills/luwu-semantic-drift-review/SKILL.md

@@ -3,8 +3,8 @@ name: luwu-semantic-drift-review
 description: >-
   Use when reviewing Luwu drift detection, rendering, normalization,
   formatting, parsers, structured comparison, or reverse synchronization. Test
-  whether the implementation distinguishes meaningful configuration changes from
-  presentation noise without changing ownership. Do not use for generic
+  whether the implementation distinguishes meaningful configuration changes
+  from presentation noise without changing ownership. Do not use for generic
   formatter or parser implementation review without a drift question.
 ---
 
@@ -45,3 +45,33 @@ Report the equivalence contract, the authority it preserves, path/line and test
 evidence, the convenience traded away, and the minimum corrective change. Mark
 implemented, partial, design-only, or unstarted. Use `docs/product.md` for the
 semantic intent and `docs/reference.md` for current comparison states.
+
+## Scope boundary
+
+This stance owns meaning versus representation: equivalence relations,
+normalization, parser subsets, formatting noise, information loss, and
+semantic reverse-sync edits. It does not own who is authorized to change a
+field, whether a mutation was confirmed, filesystem safety, sensitive-value
+exposure, or documentation freshness. Route those findings to the ownership,
+consent, boundary, confidentiality, or verifiability stance.
+
+## Good patterns
+
+- Define a narrow strict-JSON equivalence relation that preserves object
+  members and array order, distinguishes booleans from numbers, and reports
+  unsupported syntax as blocked. Evidence: the M2 contract in
+  `docs/reference.md`.
+- Patch only selected literal-JSON spans so unselected and undeclared bytes are
+  preserved. Evidence:
+  [M3 repair record](../../../docs/milestones/m3-repair-plan.md).
+
+## Bad patterns
+
+- Re-encode a complete source object to update one selected field. This can
+  silently change whitespace, escaping, number spelling, or undeclared
+  content.
+- Treat a generic formatter or parser as permission to discard unsupported
+  syntax or to call a semantically different document in sync.
+
+Patterns describe comparison risks. Parser names, supported syntax, and public
+status values remain owned by the reference contract.

.agents/skills/luwu-verifiability-review/SKILL.md

@@ -47,3 +47,32 @@ Return a maturity matrix of implemented, partial, design-only, and unstarted
 claims, with path/line evidence, missing contract or test, and the smallest
 next practice. Use `docs/status.md` for current facts, `docs/reference.md` for
 stable contracts, and `docs/product.md` for future intent.
+
+## Scope boundary
+
+This stance owns evidence continuity: whether a claim is current, planned,
+historical, testable, recoverable, and routed to one documentation owner. It
+does not own the product value itself, public field semantics, implementation
+mechanics, filesystem safety, secret exposure, consent, or configuration
+authority. Route those findings to the product, reference, design, boundary,
+confidentiality, consent, or ownership owner rather than restating them.
+
+## Good patterns
+
+- Keep current status as a small snapshot and retain dated test counts,
+  temporary paths, and closure reasoning in milestone records. Evidence:
+  [M4 closure record](../../../docs/milestones/m4.md) and
+  [current status](../../../docs/status.md).
+- Separate target replacement facts from journal publication facts and expose
+  enough metadata for the next action without persisting content. Evidence:
+  [M3 execution closure](../../../docs/milestones/m3-execution-closure.md).
+
+## Bad patterns
+
+- Copy a closed milestone's test log and temporary artifact paths into the
+  current status page, creating two competing histories.
+- Let a milestone, skill, or summary page redefine a stable field, state, or
+  error instead of linking to `docs/reference.md`.
+
+Patterns are evidence about maintainability. The status snapshot, public
+contract, and historical records keep their own document owners.