nccgs 1.1.0 → 1.3.0

package/.claude/skills/compatibility-review/SKILL.md

@@ -1,14 +1,15 @@
 ---
 name: compatibility-review
 description: Reviews save, serialization, API, network, content, platform, or migration compatibility and defines versioning, detection, fallback, tests, and rollback.
-allowed-tools: Read, Glob, Grep, Bash, Task
+allowed-tools: Read, Glob, Grep, Bash, Agent
 ---
 
 # Compatibility Review
 
-Read `.nccgs/project.yaml`, `.nccgs/state.md`, and the active context packet when present. Preserve Product Owner authority and unrelated work.
+Read models.profile in .nccgs/project.yaml; if absent, use the defaultProfile in
+.claude/nccgs/pipeline.json. Load exactly one procedure:
 
-Remain read-only unless a migration plan is requested. Inventory producers, consumers, stored/on-wire formats, versions, invariants, and failure modes. Identify breaking and ambiguous changes, old-data coverage, forward/backward expectations, migration idempotency, recovery, and telemetry. Return compatibility verdict, required tests, rollout/rollback, and Product Owner decisions.
-
-Return outcome, evidence, affected records, blockers or decisions, residual risk, and the exact next workflow. Do not claim DONE unless this is the `closure` skill and every configured gate passes.
+- astra-claude: [Astra/Claude procedure](references/astra.md).
+- quality, balanced, inherit: [Legacy procedure](references/legacy.md).
 
+Do not load or merge the other branch. Unknown profiles are configuration errors.

package/.claude/skills/compatibility-review/references/astra.md

@@ -0,0 +1,17 @@
+# Compatibility Review — astra-claude
+
+Read [the pipeline protocol](../../../nccgs/protocols/astra-claude.md) and
+[generated routing](../../../nccgs/routing.generated.md). Use Astra's supplied plan,
+requirements, and exact context references. Never substitute a legacy procedure.
+
+Domain objective: Reviews save, serialization, API, network, content, platform, or migration compatibility and defines versioning, detection, fallback, tests, and rollback.
+
+Work only inside the supplied plan and assigned write set. Activate an optional
+specialist with the agents command only when its domain is required. Keep one writer
+per path and use official Unity Skills for Editor/assets/runtime operations.
+
+This procedure is read-only: return evidence-backed findings and decisions to Astra
+or the implementer. Supplemental analysis does not replace the mandatory Fable gate.
+
+Use [task gate commands](../../../nccgs/protocols/task-gates.md) for evidence and
+state transitions. Report outcome, evidence, pending decisions, and next owner.

package/.claude/skills/compatibility-review/references/legacy.md

@@ -0,0 +1,7 @@
+# Compatibility Review
+
+Read `.nccgs/project.yaml`, `.nccgs/state.md`, and the active context packet when present. Preserve Product Owner authority and unrelated work.
+
+Remain read-only unless a migration plan is requested. Inventory producers, consumers, stored/on-wire formats, versions, invariants, and failure modes. Identify breaking and ambiguous changes, old-data coverage, forward/backward expectations, migration idempotency, recovery, and telemetry. Return compatibility verdict, required tests, rollout/rollback, and Product Owner decisions.
+
+Return outcome, evidence, affected records, blockers or decisions, residual risk, and the exact next workflow. Do not claim DONE unless this is the `closure` skill and every configured gate passes.

package/.claude/skills/context-pack/SKILL.md

@@ -6,9 +6,10 @@ allowed-tools: Read, Glob, Grep, Write, Edit
 
 # Context Pack
 
-Read `.nccgs/project.yaml`, `.nccgs/state.md`, and the active context packet when present. Preserve Product Owner authority and unrelated work.
+Read models.profile in .nccgs/project.yaml; if absent, use the defaultProfile in
+.claude/nccgs/pipeline.json. Load exactly one procedure:
 
-Read the [context packet protocol](../../nccgs/protocols/context-packets.md) and project policy. Resolve the feature ID, objective, classification, domains, and accountable owner. Map exact canonical sections and decision summaries with source revisions; list writable, read-only, protected, and generated paths; record dependencies, constraints, out-of-scope behavior, verification, and PO gates. Create or update `.nccgs/context/<feature-id>.yaml` from the template. Do not paste whole documents. Mark contradictions and the exact reason for any required full-source read. Report packet path and unresolved inputs.
-
-Return outcome, evidence, affected records, blockers or decisions, residual risk, and the exact next workflow. Do not claim DONE unless this is the `closure` skill and every configured gate passes.
+- astra-claude: [Astra/Claude procedure](references/astra.md).
+- quality, balanced, inherit: [Legacy procedure](references/legacy.md).
 
+Do not load or merge the other branch. Unknown profiles are configuration errors.

package/.claude/skills/context-pack/references/astra.md

@@ -0,0 +1,12 @@
+# Context Pack — astra-claude
+
+Read [the pipeline protocol](../../../nccgs/protocols/astra-claude.md) and
+[generated routing](../../../nccgs/routing.generated.md). Use Astra's supplied plan,
+requirements, and exact context references. Never substitute a legacy procedure.
+
+Astra owns this workflow and its canonical outputs. Claude only inspects relevant
+inputs and returns factual constraints, missing information, or proposed amendments.
+Do not create a replacement plan/document or sign off as Astra.
+
+Use [task gate commands](../../../nccgs/protocols/task-gates.md) for evidence and
+state transitions. Report outcome, evidence, pending decisions, and next owner.

package/.claude/skills/context-pack/references/legacy.md

@@ -0,0 +1,7 @@
+# Context Pack
+
+Read `.nccgs/project.yaml`, `.nccgs/state.md`, and the active context packet when present. Preserve Product Owner authority and unrelated work.
+
+Read the [context packet protocol](../../../nccgs/protocols/context-packets.md) and project policy. Resolve the feature ID, objective, classification, domains, and accountable owner. Map exact canonical sections and decision summaries with source revisions; list writable, read-only, protected, and generated paths; record dependencies, constraints, out-of-scope behavior, verification, and PO gates. Create or update `.nccgs/context/<feature-id>.yaml` from the template. Do not paste whole documents. Mark contradictions and the exact reason for any required full-source read. Report packet path and unresolved inputs.
+
+Return outcome, evidence, affected records, blockers or decisions, residual risk, and the exact next workflow. Do not claim DONE unless this is the `closure` skill and every configured gate passes.

package/.claude/skills/dependency-review/SKILL.md

@@ -1,14 +1,15 @@
 ---
 name: dependency-review
 description: Evaluates a proposed Unity package, SDK, service, or external dependency for necessity, fit, licensing, security, maintenance, compatibility, lock-in, migration, and rollback before adoption.
-allowed-tools: Read, Glob, Grep, Bash, Task, WebSearch, WebFetch
+allowed-tools: Read, Glob, Grep, Bash, Agent, WebSearch, WebFetch
 ---
 
 # Dependency Review
 
-Read `.nccgs/project.yaml`, `.nccgs/state.md`, and the active context packet when present. Preserve Product Owner authority and unrelated work.
+Read models.profile in .nccgs/project.yaml; if absent, use the defaultProfile in
+.claude/nccgs/pipeline.json. Load exactly one procedure:
 
-Remain read-only. Define the capability need and compare build, buy, existing-project, and no-change options. Verify current authoritative documentation, supported Unity/platform versions, license, data/privacy, operational cost, update policy, failure behavior, transitive risk, and exit plan. Recommend with evidence. Adoption or replacement requires explicit Product Owner authority and an architecture decision.
-
-Return outcome, evidence, affected records, blockers or decisions, residual risk, and the exact next workflow. Do not claim DONE unless this is the `closure` skill and every configured gate passes.
+- astra-claude: [Astra/Claude procedure](references/astra.md).
+- quality, balanced, inherit: [Legacy procedure](references/legacy.md).
 
+Do not load or merge the other branch. Unknown profiles are configuration errors.

package/.claude/skills/dependency-review/references/astra.md

@@ -0,0 +1,17 @@
+# Dependency Review — astra-claude
+
+Read [the pipeline protocol](../../../nccgs/protocols/astra-claude.md) and
+[generated routing](../../../nccgs/routing.generated.md). Use Astra's supplied plan,
+requirements, and exact context references. Never substitute a legacy procedure.
+
+Domain objective: Evaluates a proposed Unity package, SDK, service, or external dependency for necessity, fit, licensing, security, maintenance, compatibility, lock-in, migration, and rollback before adoption.
+
+Work only inside the supplied plan and assigned write set. Activate an optional
+specialist with the agents command only when its domain is required. Keep one writer
+per path and use official Unity Skills for Editor/assets/runtime operations.
+
+This procedure is read-only: return evidence-backed findings and decisions to Astra
+or the implementer. Supplemental analysis does not replace the mandatory Fable gate.
+
+Use [task gate commands](../../../nccgs/protocols/task-gates.md) for evidence and
+state transitions. Report outcome, evidence, pending decisions, and next owner.

package/.claude/skills/dependency-review/references/legacy.md

@@ -0,0 +1,7 @@
+# Dependency Review
+
+Read `.nccgs/project.yaml`, `.nccgs/state.md`, and the active context packet when present. Preserve Product Owner authority and unrelated work.
+
+Remain read-only. Define the capability need and compare build, buy, existing-project, and no-change options. Verify current authoritative documentation, supported Unity/platform versions, license, data/privacy, operational cost, update policy, failure behavior, transitive risk, and exit plan. Recommend with evidence. Adoption or replacement requires explicit Product Owner authority and an architecture decision.
+
+Return outcome, evidence, affected records, blockers or decisions, residual risk, and the exact next workflow. Do not claim DONE unless this is the `closure` skill and every configured gate passes.

package/.claude/skills/design/SKILL.md

@@ -1,20 +1,15 @@
 ---
 name: design
 description: Designs or revises player-facing mechanics, systems, UX, levels, economy, narrative, content, or accessibility before implementation. Routes to focused design roles and preserves Product Owner control of subjective and canonical decisions.
-allowed-tools: Read, Glob, Grep, Write, Edit, Task, AskUserQuestion, WebSearch, WebFetch
+allowed-tools: Read, Glob, Grep, Write, Edit, Agent, AskUserQuestion, WebSearch, WebFetch
 ---
 
 # Design
 
-Produce implementation-ready design truth with the smallest complete set of specialized perspectives.
+Read models.profile in .nccgs/project.yaml; if absent, use the defaultProfile in
+.claude/nccgs/pipeline.json. Load exactly one procedure:
 
-Read policy, state, relevant pillars/canon, interacting requirements, and the code constraints needed for feasibility. Distinguish exploration, focused design, and CONTROLLED cross-system change.
-
-Select the accountable design role and only the specialists required by the domain: systems, level, economy, UX, accessibility, narrative, art, audio, live ops, or prototyping. Use Opus leadership only for cross-creative conflict or product-wide synthesis.
-
-Separate established facts from proposals and Product Owner decisions. Batch meaningful choices with real alternatives, tradeoffs, and a recommendation. Do not ask implementation trivia.
-
-Define applicable rules, states, formulas, feedback, dependencies, edge cases, tuning surfaces, objective acceptance criteria, subjective gates, out-of-scope behavior, and traceable requirement IDs. Consult technical roles for feasibility without letting implementation redefine intended design.
-
-Persist accepted changes in mapped canonical sources; do not create a parallel GDD tree. Run `/design-review` for STANDARD or CONTROLLED changes. If implementation is included and decisions are closed, hand the accepted contract to `/work`.
+- astra-claude: [Astra/Claude procedure](references/astra.md).
+- quality, balanced, inherit: [Legacy procedure](references/legacy.md).
 
+Do not load or merge the other branch. Unknown profiles are configuration errors.

package/.claude/skills/design/references/astra.md

@@ -0,0 +1,12 @@
+# Design — astra-claude
+
+Read [the pipeline protocol](../../../nccgs/protocols/astra-claude.md) and
+[generated routing](../../../nccgs/routing.generated.md). Use Astra's supplied plan,
+requirements, and exact context references. Never substitute a legacy procedure.
+
+Astra owns this workflow and its canonical outputs. Claude only inspects relevant
+inputs and returns factual constraints, missing information, or proposed amendments.
+Do not create a replacement plan/document or sign off as Astra.
+
+Use [task gate commands](../../../nccgs/protocols/task-gates.md) for evidence and
+state transitions. Report outcome, evidence, pending decisions, and next owner.

package/.claude/skills/design/references/legacy.md

@@ -0,0 +1,13 @@
+# Design
+
+Produce implementation-ready design truth with the smallest complete set of specialized perspectives.
+
+Read policy, state, relevant pillars/canon, interacting requirements, and the code constraints needed for feasibility. Distinguish exploration, focused design, and CONTROLLED cross-system change.
+
+Select the accountable design role and only the specialists required by the domain: systems, level, economy, UX, accessibility, narrative, art, audio, live ops, or prototyping. Use Opus leadership only for cross-creative conflict or product-wide synthesis.
+
+Separate established facts from proposals and Product Owner decisions. Batch meaningful choices with real alternatives, tradeoffs, and a recommendation. Do not ask implementation trivia.
+
+Define applicable rules, states, formulas, feedback, dependencies, edge cases, tuning surfaces, objective acceptance criteria, subjective gates, out-of-scope behavior, and traceable requirement IDs. Consult technical roles for feasibility without letting implementation redefine intended design.
+
+Persist accepted changes in mapped canonical sources; do not create a parallel GDD tree. Run `/design-review` for STANDARD or CONTROLLED changes. If implementation is included and decisions are closed, hand the accepted contract to `/work`.

package/.claude/skills/design-review/SKILL.md

@@ -1,14 +1,15 @@
 ---
 name: design-review
 description: Reviews a proposed or changed game design for clarity, player intent, systemic interactions, balance, UX, accessibility, canon, feasibility, and testability without approving subjective choices.
-allowed-tools: Read, Glob, Grep, Task
+allowed-tools: Read, Glob, Grep, Agent
 ---
 
 # Design Review
 
-Read `.nccgs/project.yaml`, `.nccgs/state.md`, and the active context packet when present. Preserve Product Owner authority and unrelated work.
+Read models.profile in .nccgs/project.yaml; if absent, use the defaultProfile in
+.claude/nccgs/pipeline.json. Load exactly one procedure:
 
-Remain read-only. Review the target against pillars, canonical decisions, interacting systems, platform constraints, and implementation evidence only where relevant. Use focused specialists for economy, UX, accessibility, level, narrative, art, or audio lenses. Separate contradictions, exploitable/degenerate behavior, untestable requirements, and optional improvements. Return prioritized findings, PO decisions, and implementation-readiness verdict.
-
-Return outcome, evidence, affected records, blockers or decisions, residual risk, and the exact next workflow. Do not claim DONE unless this is the `closure` skill and every configured gate passes.
+- astra-claude: [Astra/Claude procedure](references/astra.md).
+- quality, balanced, inherit: [Legacy procedure](references/legacy.md).
 
+Do not load or merge the other branch. Unknown profiles are configuration errors.

package/.claude/skills/design-review/references/astra.md

@@ -0,0 +1,17 @@
+# Design Review — astra-claude
+
+Read [the pipeline protocol](../../../nccgs/protocols/astra-claude.md) and
+[generated routing](../../../nccgs/routing.generated.md). Use Astra's supplied plan,
+requirements, and exact context references. Never substitute a legacy procedure.
+
+Domain objective: Reviews a proposed or changed game design for clarity, player intent, systemic interactions, balance, UX, accessibility, canon, feasibility, and testability without approving subjective choices.
+
+Work only inside the supplied plan and assigned write set. Activate an optional
+specialist with the agents command only when its domain is required. Keep one writer
+per path and use official Unity Skills for Editor/assets/runtime operations.
+
+This procedure is read-only: return evidence-backed findings and decisions to Astra
+or the implementer. Supplemental analysis does not replace the mandatory Fable gate.
+
+Use [task gate commands](../../../nccgs/protocols/task-gates.md) for evidence and
+state transitions. Report outcome, evidence, pending decisions, and next owner.

package/.claude/skills/design-review/references/legacy.md

@@ -0,0 +1,7 @@
+# Design Review
+
+Read `.nccgs/project.yaml`, `.nccgs/state.md`, and the active context packet when present. Preserve Product Owner authority and unrelated work.
+
+Remain read-only. Review the target against pillars, canonical decisions, interacting systems, platform constraints, and implementation evidence only where relevant. Use focused specialists for economy, UX, accessibility, level, narrative, art, or audio lenses. Separate contradictions, exploitable/degenerate behavior, untestable requirements, and optional improvements. Return prioritized findings, PO decisions, and implementation-readiness verdict.
+
+Return outcome, evidence, affected records, blockers or decisions, residual risk, and the exact next workflow. Do not claim DONE unless this is the `closure` skill and every configured gate passes.

package/.claude/skills/evidence-review/SKILL.md

@@ -1,14 +1,15 @@
 ---
 name: evidence-review
 description: Independently evaluates whether accumulated tests, runtime observations, builds, reports, and manual gates actually prove each requirement and support the claimed closure state.
-allowed-tools: Read, Glob, Grep, Bash, Task
+allowed-tools: Read, Glob, Grep, Bash, Agent
 ---
 
 # Evidence Review
 
-Read `.nccgs/project.yaml`, `.nccgs/state.md`, and the active context packet when present. Preserve Product Owner authority and unrelated work.
+Read models.profile in .nccgs/project.yaml; if absent, use the defaultProfile in
+.claude/nccgs/pipeline.json. Load exactly one procedure:
 
-Read the [evidence protocol](../../nccgs/protocols/evidence.md), requirements, evidence records, test outputs, runtime/build observations, baseline, and waivers. Map every criterion to PASS, FAIL, NOT PROVEN, PO GATE, or N/A. Detect stale source revisions, unexplained count/skip changes, evidence produced before the final diff, and claims based only on inspection. Return closure impact and exact evidence still required.
-
-Return outcome, evidence, affected records, blockers or decisions, residual risk, and the exact next workflow. Do not claim DONE unless this is the `closure` skill and every configured gate passes.
+- astra-claude: [Astra/Claude procedure](references/astra.md).
+- quality, balanced, inherit: [Legacy procedure](references/legacy.md).
 
+Do not load or merge the other branch. Unknown profiles are configuration errors.

package/.claude/skills/evidence-review/references/astra.md

@@ -0,0 +1,17 @@
+# Evidence Review — astra-claude
+
+Read [the pipeline protocol](../../../nccgs/protocols/astra-claude.md) and
+[generated routing](../../../nccgs/routing.generated.md). Use Astra's supplied plan,
+requirements, and exact context references. Never substitute a legacy procedure.
+
+Domain objective: Independently evaluates whether accumulated tests, runtime observations, builds, reports, and manual gates actually prove each requirement and support the claimed closure state.
+
+Work only inside the supplied plan and assigned write set. Activate an optional
+specialist with the agents command only when its domain is required. Keep one writer
+per path and use official Unity Skills for Editor/assets/runtime operations.
+
+This procedure is read-only: return evidence-backed findings and decisions to Astra
+or the implementer. Supplemental analysis does not replace the mandatory Fable gate.
+
+Use [task gate commands](../../../nccgs/protocols/task-gates.md) for evidence and
+state transitions. Report outcome, evidence, pending decisions, and next owner.

package/.claude/skills/evidence-review/references/legacy.md

@@ -0,0 +1,7 @@
+# Evidence Review
+
+Read `.nccgs/project.yaml`, `.nccgs/state.md`, and the active context packet when present. Preserve Product Owner authority and unrelated work.
+
+Read the [evidence protocol](../../../nccgs/protocols/evidence.md), requirements, evidence records, test outputs, runtime/build observations, baseline, and waivers. Map every criterion to PASS, FAIL, NOT PROVEN, PO GATE, or N/A. Detect stale source revisions, unexplained count/skip changes, evidence produced before the final diff, and claims based only on inspection. Return closure impact and exact evidence still required.
+
+Return outcome, evidence, affected records, blockers or decisions, residual risk, and the exact next workflow. Do not claim DONE unless this is the `closure` skill and every configured gate passes.

package/.claude/skills/hotfix/SKILL.md

@@ -1,14 +1,15 @@
 ---
 name: hotfix
 description: Runs a tightly scoped CONTROLLED workflow for urgent production or release-critical defects with containment, minimal change, accelerated evidence, rollback, and follow-up debt.
-allowed-tools: Read, Glob, Grep, Write, Edit, Bash, Task, AskUserQuestion
+allowed-tools: Read, Glob, Grep, Write, Edit, Bash, Agent, AskUserQuestion
 ---
 
 # Hotfix
 
-Read `.nccgs/project.yaml`, `.nccgs/state.md`, and the active context packet when present. Preserve Product Owner authority and unrelated work.
+Read models.profile in .nccgs/project.yaml; if absent, use the defaultProfile in
+.claude/nccgs/pipeline.json. Load exactly one procedure:
 
-Confirm affected version, severity, user/data impact, reproduction, containment, release branch, and rollback. Build the smallest safe fix with one accountable writer; preserve compatibility and avoid unrelated refactors. Run targeted regression plus the release-critical checks justified by blast radius. Record evidence, known gaps, approval, release steps, rollback trigger, and required follow-up. Publication still requires explicit authority.
-
-Return outcome, evidence, affected records, blockers or decisions, residual risk, and the exact next workflow. Do not claim DONE unless this is the `closure` skill and every configured gate passes.
+- astra-claude: [Astra/Claude procedure](references/astra.md).
+- quality, balanced, inherit: [Legacy procedure](references/legacy.md).
 
+Do not load or merge the other branch. Unknown profiles are configuration errors.

package/.claude/skills/hotfix/references/astra.md

@@ -0,0 +1,23 @@
+# Hotfix — astra-claude
+
+Read [the pipeline protocol](../../../nccgs/protocols/astra-claude.md) and
+[generated routing](../../../nccgs/routing.generated.md). Use Astra's supplied plan,
+requirements, and exact context references. Never substitute a legacy procedure.
+
+Domain objective: Runs a tightly scoped CONTROLLED workflow for urgent production or release-critical defects with containment, minimal change, accelerated evidence, rollback, and follow-up debt.
+
+Work only inside the supplied plan and assigned write set. Activate an optional
+specialist with the agents command only when its domain is required. Keep one writer
+per path and use official Unity Skills for Editor/assets/runtime operations.
+
+For code changes: task init with Astra's plan, bind the session, implement and test,
+record implementation evidence, request independent review, fix and re-review as
+needed, then task check --target report. Only after that check passes write and
+record the report for Astra. Stop at AWAITING_ASTRA_CHECK. Astra checking and the
+Product Owner's explicit acceptance happen outside the Claude implementation session.
+
+Publication and other external release actions still require explicit Product Owner
+authority; successful local verification grants no publication permission.
+
+Use [task gate commands](../../../nccgs/protocols/task-gates.md) for evidence and
+state transitions. Report outcome, evidence, pending decisions, and next owner.

package/.claude/skills/hotfix/references/legacy.md

@@ -0,0 +1,7 @@
+# Hotfix
+
+Read `.nccgs/project.yaml`, `.nccgs/state.md`, and the active context packet when present. Preserve Product Owner authority and unrelated work.
+
+Confirm affected version, severity, user/data impact, reproduction, containment, release branch, and rollback. Build the smallest safe fix with one accountable writer; preserve compatibility and avoid unrelated refactors. Run targeted regression plus the release-critical checks justified by blast radius. Record evidence, known gaps, approval, release steps, rollback trigger, and required follow-up. Publication still requires explicit authority.
+
+Return outcome, evidence, affected records, blockers or decisions, residual risk, and the exact next workflow. Do not claim DONE unless this is the `closure` skill and every configured gate passes.

package/.claude/skills/incident-recovery/SKILL.md

@@ -1,14 +1,15 @@
 ---
 name: incident-recovery
 description: Coordinates evidence-preserving response to data loss, corruption, broken builds, failed migrations, or production-impacting Unity incidents.
-allowed-tools: Read, Glob, Grep, Write, Edit, Bash, Task, AskUserQuestion
+allowed-tools: Read, Glob, Grep, Write, Edit, Bash, Agent, AskUserQuestion
 ---
 
 # Incident Recovery
 
-Read `.nccgs/project.yaml`, `.nccgs/state.md`, and the active context packet when present. Preserve Product Owner authority and unrelated work.
+Read models.profile in .nccgs/project.yaml; if absent, use the defaultProfile in
+.claude/nccgs/pipeline.json. Load exactly one procedure:
 
-Prioritize containment and evidence preservation. Record incident scope, affected versions/data, current state, timeline, hypotheses, and irreversible-action boundaries. Avoid destructive cleanup and speculative rewrites. Build a recovery graph with backups, reproducible diagnostics, validation, and rollback. Apply only authorized recovery changes; maintain an incident record and follow-up prevention actions. Data deletion or production actions require explicit authority.
-
-Return outcome, evidence, affected records, blockers or decisions, residual risk, and the exact next workflow. Do not claim DONE unless this is the `closure` skill and every configured gate passes.
+- astra-claude: [Astra/Claude procedure](references/astra.md).
+- quality, balanced, inherit: [Legacy procedure](references/legacy.md).
 
+Do not load or merge the other branch. Unknown profiles are configuration errors.

package/.claude/skills/incident-recovery/references/astra.md

@@ -0,0 +1,20 @@
+# Incident Recovery — astra-claude
+
+Read [the pipeline protocol](../../../nccgs/protocols/astra-claude.md) and
+[generated routing](../../../nccgs/routing.generated.md). Use Astra's supplied plan,
+requirements, and exact context references. Never substitute a legacy procedure.
+
+Domain objective: Coordinates evidence-preserving response to data loss, corruption, broken builds, failed migrations, or production-impacting Unity incidents.
+
+Work only inside the supplied plan and assigned write set. Activate an optional
+specialist with the agents command only when its domain is required. Keep one writer
+per path and use official Unity Skills for Editor/assets/runtime operations.
+
+For code changes: task init with Astra's plan, bind the session, implement and test,
+record implementation evidence, request independent review, fix and re-review as
+needed, then task check --target report. Only after that check passes write and
+record the report for Astra. Stop at AWAITING_ASTRA_CHECK. Astra checking and the
+Product Owner's explicit acceptance happen outside the Claude implementation session.
+
+Use [task gate commands](../../../nccgs/protocols/task-gates.md) for evidence and
+state transitions. Report outcome, evidence, pending decisions, and next owner.

package/.claude/skills/incident-recovery/references/legacy.md

@@ -0,0 +1,7 @@
+# Incident Recovery
+
+Read `.nccgs/project.yaml`, `.nccgs/state.md`, and the active context packet when present. Preserve Product Owner authority and unrelated work.
+
+Prioritize containment and evidence preservation. Record incident scope, affected versions/data, current state, timeline, hypotheses, and irreversible-action boundaries. Avoid destructive cleanup and speculative rewrites. Build a recovery graph with backups, reproducible diagnostics, validation, and rollback. Apply only authorized recovery changes; maintain an incident record and follow-up prevention actions. Data deletion or production actions require explicit authority.
+
+Return outcome, evidence, affected records, blockers or decisions, residual risk, and the exact next workflow. Do not claim DONE unless this is the `closure` skill and every configured gate passes.

package/.claude/skills/localize-game/SKILL.md

@@ -1,14 +1,15 @@
 ---
 name: localize-game
 description: Plans, implements, or reviews internationalization and localization across strings, fonts, layout, input, assets, culture, and LQA using the project's supported locales.
-allowed-tools: Read, Glob, Grep, Write, Edit, Bash, Task, AskUserQuestion, WebSearch, WebFetch
+allowed-tools: Read, Glob, Grep, Write, Edit, Bash, Agent, AskUserQuestion, WebSearch, WebFetch
 ---
 
 # Localize Game
 
-Read `.nccgs/project.yaml`, `.nccgs/state.md`, and the active context packet when present. Preserve Product Owner authority and unrelated work.
+Read models.profile in .nccgs/project.yaml; if absent, use the defaultProfile in
+.claude/nccgs/pipeline.json. Load exactly one procedure:
 
-Read locale scope, canonical terminology, UI/content sources, existing localization architecture, and platform constraints. Route writing, UI, accessibility, content, and Unity specialists as needed. Preserve stable string IDs, context, placeholders, plurals, fallback, font coverage, layout expansion, and extraction determinism. Use the official `localization` Unity Skill for Unity operation mechanics. Do not invent locale scope or approve translations without review evidence.
-
-Return outcome, evidence, affected records, blockers or decisions, residual risk, and the exact next workflow. Do not claim DONE unless this is the `closure` skill and every configured gate passes.
+- astra-claude: [Astra/Claude procedure](references/astra.md).
+- quality, balanced, inherit: [Legacy procedure](references/legacy.md).
 
+Do not load or merge the other branch. Unknown profiles are configuration errors.

package/.claude/skills/localize-game/references/astra.md

@@ -0,0 +1,20 @@
+# Localize Game — astra-claude
+
+Read [the pipeline protocol](../../../nccgs/protocols/astra-claude.md) and
+[generated routing](../../../nccgs/routing.generated.md). Use Astra's supplied plan,
+requirements, and exact context references. Never substitute a legacy procedure.
+
+Domain objective: Plans, implements, or reviews internationalization and localization across strings, fonts, layout, input, assets, culture, and LQA using the project's supported locales.
+
+Work only inside the supplied plan and assigned write set. Activate an optional
+specialist with the agents command only when its domain is required. Keep one writer
+per path and use official Unity Skills for Editor/assets/runtime operations.
+
+For code changes: task init with Astra's plan, bind the session, implement and test,
+record implementation evidence, request independent review, fix and re-review as
+needed, then task check --target report. Only after that check passes write and
+record the report for Astra. Stop at AWAITING_ASTRA_CHECK. Astra checking and the
+Product Owner's explicit acceptance happen outside the Claude implementation session.
+
+Use [task gate commands](../../../nccgs/protocols/task-gates.md) for evidence and
+state transitions. Report outcome, evidence, pending decisions, and next owner.

package/.claude/skills/localize-game/references/legacy.md

@@ -0,0 +1,7 @@
+# Localize Game
+
+Read `.nccgs/project.yaml`, `.nccgs/state.md`, and the active context packet when present. Preserve Product Owner authority and unrelated work.
+
+Read locale scope, canonical terminology, UI/content sources, existing localization architecture, and platform constraints. Route writing, UI, accessibility, content, and Unity specialists as needed. Preserve stable string IDs, context, placeholders, plurals, fallback, font coverage, layout expansion, and extraction determinism. Use the official `localization` Unity Skill for Unity operation mechanics. Do not invent locale scope or approve translations without review evidence.
+
+Return outcome, evidence, affected records, blockers or decisions, residual risk, and the exact next workflow. Do not claim DONE unless this is the `closure` skill and every configured gate passes.

package/.claude/skills/migrate-project/SKILL.md

@@ -1,23 +1,15 @@
 ---
 name: migrate-project
 description: Reviews, plans, applies, and verifies adoption or upgrade of NCCGS in an existing Unity project. Preserves canonical truth, ownership, closure safeguards, model policy, local settings, and rollback while replacing obsolete workflow machinery.
-allowed-tools: Read, Glob, Grep, Write, Edit, Bash, Task, AskUserQuestion
+allowed-tools: Read, Glob, Grep, Write, Edit, Bash, Agent, AskUserQuestion
 ---
 
 # Migrate project
 
-Treat brownfield adoption and migrations explicitly requested by the installer as CONTROLLED work. Read the complete [migration procedure](references/procedure.md). Default to `full` when no mode is given. Do not require migration merely because NCCGS was updated when the installer reported `No project migration required`.
+Read models.profile in .nccgs/project.yaml; if absent, use the defaultProfile in
+.claude/nccgs/pipeline.json. Load exactly one procedure:
 
-## Modes
+- astra-claude: [Astra/Claude procedure](references/astra.md).
+- quality, balanced, inherit: [Legacy procedure](references/legacy.md).
 
-- `full` (default): run review, plan, apply, and verify in one invocation. Continue across phase boundaries without asking for ceremonial approval. Pause only for an unresolved Product Owner decision, destructive action, external side effect, or authority outside the current request.
-- `review`: read-only inventory and KEEP/ADAPT/MIGRATE/DEPRECATE/REMOVE/MISSING classification.
-- `plan`: write a dated reversible plan and proposed schema-2 policy without overwriting project truth.
-- `apply`: execute an approved plan in validated batches with backups.
-- `verify`: prove policy, routing, hooks, Unity integration, evidence, closure, and a representative work path.
-
-Migration decisions must cover canonical sources, ownership/protected paths, architecture, Unity tooling, model profile/fallback, enabled departments, context packets, verification baselines, closure safeguards, Git tracking, local settings, obsolete global skills, and rollback.
-
-In `full` mode, write the plan before applying it, preserve the same evidence and rollback guarantees as the individual modes, and stop safely if review finds a decision that would materially change project truth. A reversible plan created within the same invocation is approved for in-scope application unless project policy explicitly requires a separate approval.
-
-Never remove reports, registers, checklists, user settings, source assets, or legacy instructions until their live purpose is mapped and the Product Owner approves the replacement. Do not infer canon from code. Return batches, commits, preserved evidence, unresolved decisions, rollback, and final migration state.
+Do not load or merge the other branch. Unknown profiles are configuration errors.

package/.claude/skills/migrate-project/references/astra.md

@@ -0,0 +1,23 @@
+# Migrate project — astra-claude
+
+Read [the pipeline protocol](../../../nccgs/protocols/astra-claude.md) and
+[generated routing](../../../nccgs/routing.generated.md). Use Astra's supplied plan,
+requirements, and exact context references. Never substitute a legacy procedure.
+
+Domain objective: Reviews, plans, applies, and verifies adoption or upgrade of NCCGS in an existing Unity project. Preserves canonical truth, ownership, closure safeguards, model policy, local settings, and rollback while replacing obsolete workflow machinery.
+
+Work only inside the supplied plan and assigned write set. Activate an optional
+specialist with the agents command only when its domain is required. Keep one writer
+per path and use official Unity Skills for Editor/assets/runtime operations.
+
+For code changes: task init with Astra's plan, bind the session, implement and test,
+record implementation evidence, request independent review, fix and re-review as
+needed, then task check --target report. Only after that check passes write and
+record the report for Astra. Stop at AWAITING_ASTRA_CHECK. Astra checking and the
+Product Owner's explicit acceptance happen outside the Claude implementation session.
+
+Default to full inventory/apply/verify using Astra's existing migration plan.
+Preserve project overrides and backups; missing planning inputs go back to Astra.
+
+Use [task gate commands](../../../nccgs/protocols/task-gates.md) for evidence and
+state transitions. Report outcome, evidence, pending decisions, and next owner.

package/.claude/skills/migrate-project/references/legacy.md

@@ -0,0 +1,19 @@
+# Migrate project
+
+Treat brownfield adoption and migrations explicitly requested by the installer as CONTROLLED work. Read the complete [migration procedure](../references/procedure.md). Default to `full` when no mode is given. Do not require migration merely because NCCGS was updated when the installer reported `No project migration required`.
+
+When `.nccgs/framework-overrides.json` exists, treat every `pending_review` entry as preserved project truth, not installer debris. Compare the project version with the current NCCGS version, keep or rebase the customization deliberately, and record the decision in the migration plan. Only after every pending item has been reviewed and the final bytes or intentional absence are correct, run `node .claude/nccgs/tools/approve-overrides.mjs --project . --approve-pending`. Never approve first to make `doctor` green.
+
+## Modes
+
+- `full` (default): run review, plan, apply, and verify in one invocation. Continue across phase boundaries without asking for ceremonial approval. Pause only for an unresolved Product Owner decision, destructive action, external side effect, or authority outside the current request.
+- `review`: read-only inventory and KEEP/ADAPT/MIGRATE/DEPRECATE/REMOVE/MISSING classification.
+- `plan`: write a dated reversible plan and proposed schema-2 policy without overwriting project truth.
+- `apply`: execute an approved plan in validated batches with backups.
+- `verify`: prove policy, routing, hooks, Unity integration, evidence, closure, and a representative work path.
+
+Migration decisions must cover canonical sources, ownership/protected paths, architecture, Unity tooling, model profile/fallback, enabled departments, context packets, verification baselines, closure safeguards, Git tracking, local settings, preserved framework overrides, obsolete global skills, and rollback.
+
+In `full` mode, write the plan before applying it, preserve the same evidence and rollback guarantees as the individual modes, and stop safely if review finds a decision that would materially change project truth. A reversible plan created within the same invocation is approved for in-scope application unless project policy explicitly requires a separate approval.
+
+Never remove reports, registers, checklists, user settings, source assets, or legacy instructions until their live purpose is mapped and the Product Owner approves the replacement. Do not infer canon from code. Return batches, commits, preserved evidence, unresolved decisions, rollback, and final migration state.

package/.claude/skills/migrate-project/references/procedure.md

@@ -28,6 +28,10 @@ requires no project migration, report that fact and do not manufacture migration
    strongest active safeguard until the PO explicitly accepts a weaker model.
 7. Determine the project model profile, enabled departments, permitted fallback,
    orchestration depth/parallelism, context-packet locations, and hook compatibility.
+8. If `.nccgs/framework-overrides.json` exists, inspect every override. Diff its
+   preserved bytes or intentional absence against the installed upstream version,
+   and distinguish a still-needed project policy from a customization now supplied
+   by NCCGS. Line-ending-only differences are not overrides.
 
 ## Plan
 
@@ -42,6 +46,7 @@ Create `.nccgs/migrations/<date>-migration-plan.md` with:
 - ordered reversible batches and dependencies;
 - exact broad/irreversible decisions requiring PO authority;
 - backup/rollback method;
+- an explicit KEEP/REBASE/RETIRE decision for every pending framework override;
 - validation for each batch and final end-to-end proof.
 
 Prioritize authority and policy conflicts, then model/routing conflicts, Unity
@@ -64,6 +69,12 @@ traceability/evidence, lifecycle hooks, workflow replacement, and cleanup.
 - Do not leave two active agents/skills advertising mutually exclusive Unity
   execution bridges. Disable obsolete global routing signals outside the repository
   through an explicit user action.
+- Preserve pending framework overrides while reviewing them. Rebase each retained
+  override onto the newly installed upstream content when needed. After all entries
+  match the migration decisions, run
+  `node .claude/nccgs/tools/approve-overrides.mjs --project . --approve-pending`.
+  Approval records the final hashes; it is the last apply step, never a substitute
+  for the review.
 
 ## Verify
 
@@ -72,4 +83,13 @@ role/workflow catalogs, bounded delegation, hook health, context packet freshnes
 canonical map, requirements/state, Unity Skills integrity, CLI/Pipeline operation,
 relevant tests, runtime evidence, and subjective-gate routing. Verify that required
 linked artifacts block DONE in `linked` or `strict` mode. Use a small representative
-task or dry run when product changes are not authorized.
+task or dry run when product changes are not authorized. Run `npx nccgs@latest doctor`
+and require zero pending framework overrides before declaring migration complete.
+
+## Astra-owned planning exception
+
+Under astra-claude, Astra owns the migration plan, policy decisions, and canonical
+document changes. Claude may inventory and implement a supplied migration plan;
+full must pause at missing Astra inputs instead of creating a replacement plan.
+Code changes require Opus 5.5/xhigh, Fable review, a report to Astra, Astra checking,
+and explicit Product Owner acceptance before DONE. See the profile handoff protocol.