nccgs 2.1.0 → 2.1.1

package/MEMORY.md

@@ -1,60 +1,109 @@
-# NCCGS — portable framework decisions
-
-NCCGS means **New Claude Code Game Studio** and targets **Claude Code**.
-This is the unpublished 2.1.0 source, adding reversible project uninstallation.
-The earlier 2.0.0 release was published on 2026-10-28; its archive is unchanged.
-Local preparation does not change npm tags. Native qualification remains scoped
-to its frozen archives; see README.md and CHANGELOG.md for evidence boundaries.
-Project-specific authorizations, provider sessions, budgets and internal development
-history are retained separately, outside the public package. Nothing in this file
-authorizes spending, provider access or changes to a user's project by itself.
-
-## Execution and authority
-
-- Follow the user's explicit instructions and the project's approved contract.
-- New installations use standalone Claude Code execution; external review is optional
-  and separately configured. Preserve existing execution/profile obligations on update.
-- Codex may maintain and test framework source. It must not replace Claude Code agents
-  to claim actual NCCGS game orchestration. Installed files, configured models and
-  synthetic tests are not provider execution evidence.
-- Use actual correlated native worker observations. Unknown model, effort or usage
-  remains unknown. Never infer execution from a requested alias or a handoff marker.
-- Preserve failed attempts, negative reviews, task history, evidence and consumed
-  allowances. An update, migration, retry or new report must not reset spent budgets.
-- Uninstall only from an external terminal after sessions and owned work stop.
-  Preserve game source, custom files and .nccgs history/budgets. Uninstall never
-  closes tasks, waives review/acceptance or authorizes deleting retained records.
-- Known-solution or contaminated benchmark baselines remain reference-only; do not
-  score them as independent NCCGS runs.
-
-## Delivery and evidence
-
-- Evaluate brief fidelity, implementation, tests, integration, review/repair and
-  evidence-backed handoff. Product balance and feel are separate unless the contract
-  explicitly includes them. Do not expand a delivery task into an unrequested study.
-- Keep approved gameplay rules, formulas and UI/accessibility requirements. A correct
-  loss can pass a behavior test. Disclosed fixtures may arrange preconditions but must
-  exercise real logic rather than inject the asserted outcome.
-- Automatic acceptance requires current measured assertions; helper exit zero and a
-  reviewer boolean alone do not prove success. Reject missing, stale or malformed data.
-- A completed independent reviewer owns its exact verdict/findings/conditions. PASS
-  requires no unresolved mandatory conditions; do not translate or edit its verdict.
-- Technical acceptance, human/Product Owner acceptance and benchmark claims are distinct.
-  Never claim broad performance, quality or productivity from one supported case.
-- Keep frozen evaluations unchanged when maintaining the framework or preparing a release.
-
-## Operational boundaries
-
-- Launch/resume through nccgs run. Honor native, OS, managed, destructive-action and
-  project permission constraints; do not repeatedly request already-authorized routine work.
-- Use foreground Agent calls and managed verification jobs for long engine checks.
-  Keep process ownership, finite time/cost/dispatch limits and interruption evidence.
-  Never escape ownership through a detached or WMI-launched process.
-- Task lifecycle commands are one plain invocation per shell call, without pipes,
-  redirection or chained commands. Read task state with inspect.
-- Check effective native role limits; prompt text does not change maxTurns. The
-  reviewer ceiling is 200 turns, not a target or a budget extension.
-- Use reviewed design readiness for features and exact bounded repair packets when
-  required. Read the relevant protocols rather than guessing governance syntax.
-- Preserve bundled third-party licenses and attribution, including Ponytail guidance
-  and the official Unity Skills. Additional guidance does not waive acceptance gates.
+# NCCGS — portable framework decisions
+
+NCCGS means **New Claude Code Game Studio** and targets **Claude Code**.
+This is the unpublished 2.1.1 release source, adding reversible project uninstallation.
+The earlier 2.0.0 release was published on 2026-10-28; its archive is unchanged.
+The registry's existing 2.1.0 artifact is distinct from the accepted local R11 r05
+baseline; 2.1.1 is its separately prepared release identity.
+Local preparation does not change npm tags. Native qualification remains scoped
+to its frozen archives; see README.md and CHANGELOG.md for evidence boundaries.
+Project-specific authorizations, provider sessions, budgets and internal development
+history are retained separately, outside the public package. Nothing in this file
+authorizes spending, provider access or changes to a user's project by itself.
+
+## Execution and authority
+
+- Follow the user's explicit instructions and the project's approved contract.
+- In the user's shared workflow, the sequence is PO + ChatGPT/Codex brainstorm,
+  confirm the approach, ChatGPT/Codex issue the English prompt, Claude Code executes
+  through NCCGS, then PO + ChatGPT/Codex review. NCCGS's entire internal workflow
+  belongs to the Claude execution step. The external role is ChatGPT/Codex; no
+  additional named planner or compatibility-role mapping is part of this contract.
+  Internal planning, workers, reviews and task states remain NCCGS responsibilities
+  within the approved assignment. Standalone execution is not itself a conflict;
+  do not require compatibility migration or native external gates just to represent
+  the outer workflow. Internal completion returns a candidate for external review.
+  Preserve actual existing gates/history until a separately authorized change.
+- New installations use standalone Claude Code execution; external review is optional
+  and separately configured. Preserve existing execution/profile obligations on update.
+- Codex may maintain and test framework source. It must not replace Claude Code agents
+  to claim actual NCCGS game orchestration. Installed files, configured models and
+  synthetic tests are not provider execution evidence.
+- Use actual correlated native worker observations. Unknown model, effort or usage
+  remains unknown. Never infer execution from a requested alias or a handoff marker.
+- Preserve failed attempts, negative reviews, task history, evidence and consumed
+  allowances. An update, migration, retry or new report must not reset spent budgets.
+- Uninstall only from an external terminal after sessions and owned work stop.
+  Preserve game source, custom files and .nccgs history/budgets. Uninstall never
+  closes tasks, waives review/acceptance or authorizes deleting retained records.
+- Known-solution or contaminated benchmark baselines remain reference-only; do not
+  score them as independent NCCGS runs.
+
+## Delivery and evidence
+
+- Evaluate brief fidelity, implementation, tests, integration, review/repair and
+  evidence-backed handoff. Product balance and feel are separate unless the contract
+  explicitly includes them. Do not expand a delivery task into an unrequested study.
+- Keep approved gameplay rules, formulas and UI/accessibility requirements. A correct
+  loss can pass a behavior test. Disclosed fixtures may arrange preconditions but must
+  exercise real logic rather than inject the asserted outcome.
+- Automatic acceptance requires current measured assertions; helper exit zero and a
+  reviewer boolean alone do not prove success. Reject missing, stale or malformed data.
+- A completed independent reviewer owns its exact verdict/findings/conditions. PASS
+  requires no unresolved mandatory conditions; do not translate or edit its verdict.
+- Technical acceptance, human/Product Owner acceptance and benchmark claims are distinct.
+  Never claim broad performance, quality or productivity from one supported case.
+- Keep frozen evaluations unchanged when maintaining the framework or preparing a release.
+
+## Operational boundaries
+
+- Launch/resume through nccgs run. Honor native, OS, managed, destructive-action and
+  project permission constraints; do not repeatedly request already-authorized routine work.
+- Use foreground Agent calls and managed verification jobs for long engine checks.
+  Keep process ownership, finite time/cost/dispatch limits and interruption evidence.
+  Never escape ownership through a detached or WMI-launched process.
+- Task lifecycle commands are one plain invocation per shell call, without pipes,
+  redirection or chained commands. Read task state with inspect.
+- Check effective native role limits; prompt text does not change maxTurns. The
+  reviewer ceiling is 200 turns, not a target or a budget extension.
+- Use reviewed design readiness for features and exact bounded repair packets when
+  required. Read the relevant protocols rather than guessing governance syntax.
+- Preserve bundled third-party licenses and attribution, including Ponytail guidance
+  and the official Unity Skills. Additional guidance does not waive acceptance gates.
+
+## Planned framework additions
+
+- Add focused asset-production support, drawing on `create-game-assets` from
+  https://github.com/gamedev-skills/awesome-gamedev-agent-skills.
+  Prioritize the art-direction brief, asset manifest, and production/import/visual-QA
+  checklists through the existing Art Direction, Technical Artist and Asset Audit
+  responsibilities. Preserve project art direction, approval authority and provenance;
+  review and retain applicable upstream licensing when adapting content.
+  This is a recorded future addition, not an implemented integration. Do not import
+  the full skills catalog, add a second router, or make image generation mandatory.
+- Add optional Graphify integration only for projects that already use it. A global
+  `graphifyy` installation alone does not activate it. Skip projects without a setup;
+  NCCGS does not install or upgrade Graphify. For usable existing setups, synchronize
+  changed indexed inputs in settled batches, including semantic document processing,
+  and preserve explicit pending/stale status on failure. Shared workflow/project
+  configuration owns the tool and corpus; NCCGS owns its change handoff and truthful
+  freshness reporting. Shared-workflow authors and NCCGS must use the same effective
+  project configuration, corpus, graph location and freshness contract. No separate
+  NCCGS graph or independently editable integration policy. Both sides account for
+  their selected file changes; one updater at a time refreshes the shared graph after
+  writers settle, before graph-dependent handoff/review snapshot capture. This
+  integration is planned, not implemented.
+- The two-feature implementation plan is
+  [Asset Production and Graphify Update Plan](docs/ASSET-PRODUCTION-GRAPHIFY-PLAN.md).
+  Archify remains outside NCCGS as an optional shared-workflow visualization tool.
+
+## Unity CLI-only transport
+
+Unity operations use direct Unity CLI and the approved com.unity.pipeline package.
+Do not install, configure, discover or call Unity MCP, including Coplay,
+AnkleBreaker or the official CLI's optional unity mcp mode. No MCP fallback is
+authorized by a licensing/connection failure. Follow the NCCGS unity-boundary
+protocol before applying generic Unity Skills; upstream MCP examples remain
+reference only. Non-Unity integrations need an explicit external configuration
+and are not removed or enabled globally by this package. Existing gates, models,
+history and consumed budgets remain authoritative.

package/README.md

@@ -1,191 +1,206 @@
-# NCCGS — New Claude Code Game Studio
-
-**2.1.0** is the unpublished next package for standalone game-development
-orchestration in **Claude Code**. It adds reversible project uninstallation;
-the published **2.0.0** archive remains unchanged. Local preparation moves no npm tags.
-It retains handoff automation, bounded task slicing and Unity repair, protected
-final-review allocation, managed Unity settlement and exact first-import output
-declarations. New repair steps block unsaved loaded scenes before execution.
-See [handoff operations](docs/HANDOFF-AUTOMATION.md), [bounded repair](docs/BOUNDED-REPAIR.md),
-[task slicing](docs/TASK-SLICING.md), [uninstallation](docs/UNINSTALL.md) and [release notes](CHANGELOG.md).
-
-Claude Code handles intake, planning, specialist dispatch, testing, independent
-review and reporting. External review can be configured separately. This major
-version introduces standalone defaults and explicit execution-policy migration
-from the older compatibility workflow; existing task obligations remain preserved.
-
-## Validation and limits
-
-The frozen exact RC4 archive was exercised through actual Claude Code on a
-controlled Unity scene repair: dirty-scene refusal preserved unsaved content,
-a fresh worker repaired the target, deterministic PARTIAL handoff required no
-continuation, all five checks and 28 criteria/385 assertions passed, and a fresh
-independent reviewer returned unconditional PASS before normal task closure.
-This is scoped workflow evidence, not broad game-quality or productivity proof.
-
-Version 2.0.0 keeps those runtime guards and adds a receipt publisher so external
-Editor stop evidence is available inside the reviewer's project. It does not
-operate processes or alter acceptance. See [Editor evidence](docs/EDITOR-SESSION-EVIDENCE.md).
-The exact 2.0.0 archive subsequently passed its scoped native Unity qualification
-and was published on 2026-10-28. Frozen release receipts retain its tests and
-independent reviews. The new 2.1.0 uninstaller receives local filesystem/CLI tests;
-that does not claim another Claude Code/Unity run. The first independent maintenance
-review required four fixes; this candidate adds those fixes and awaits qualification
-of its exact archive, including a fresh independent review. Earlier verdicts remain
-unchanged in their separate receipts.
-Frozen history remains intact.
-See [release status](docs/RELEASE-STATUS.md), [upgrading](UPGRADING.md),
-[Vietnamese quick start](docs/HUONG-DAN-MIGRATE-VA-SU-DUNG.md) and [decisions](MEMORY.md).
-
-## What changed
-
-Standalone `/design`, `/design-review` and `/story-readiness` now load the
-[design lifecycle](.claude/nccgs/protocols/design-lifecycle.md). New installs require
-a reviewed design handoff before dependent feature implementation; criteria and
-their source/threshold authority are pinned across that handoff. The fixed
-`task design-validate` command supports plan tasks without granting arbitrary
-shell execution. Existing policies and frozen evaluations retain their obligations;
-see [upgrading](UPGRADING.md). Schema validation is not semantic or gameplay acceptance.
-
-
-
-NCCGS includes a [shared Ponytail package](.claude/nccgs/packages/ponytail/README.md)
-for technical planning, implementation, supplementary complexity review and known
-code limits. The coordinator and agents load it by assignment through their common
-contracts. It borrows selected guidance from DietrichGebert/Ponytail under MIT;
-the [original license](.claude/nccgs/packages/ponytail/LICENSE) and
-[attribution/change notice](.claude/nccgs/packages/ponytail/NOTICE.md) ship with every
-installation. It preserves full design and QA obligations; no separate Ponytail
-plugin or global hooks are installed. Broad comparative effectiveness remains unmeasured.
-
-- Separate execution policy, model profile and external review (`off`, `advisory`,
-  `required`); project updates preserve existing obligations.
-- Task contracts for `implement`, `audit` and `plan`, with conservative
-  FAST/STANDARD/CONTROLLED classification, reasons, inputs and acceptance gates.
-- Durable dispatch reservations, path ownership checks, two-worker ceiling,
-  independent reviewer slot, ancestor quotas and recovery without budget refunds.
-- Final evidence tied to task, source, contract, runtime invocation and raw source.
-  Missing model/effort/usage is explicit; requested settings do not prove execution.
-- Reports for success, partial, blocked and failed outcomes; strict successful
-  closure, optional external receipts and explicit acceptance when configured.
-- Execution migration with dry-run, archived configuration and guarded rollback.
-
-The studio retains 47 roles (13 active by default, including Game Designer), 36 NCCGS workflows, scoped
-rules, lifecycle hooks and unmodified official Unity Skills. Enable specialists
-when their expertise resolves a concrete problem; do not dispatch the whole studio.
-
-## Try the local 2.1.0 package
-
-Requirements: Claude Code and Node.js 18+. Unity CLI/Pipeline is needed for live
-Unity operations. Run against a disposable Unity project first:
-
-Use the locally prepared archive and its accompanying SHA-256 manifest:
-
-```powershell
-$NccgsArchive = "D:\path\to\nccgs-2.1.0.tgz"
-npx --yes --package "$NccgsArchive" nccgs install --project "D:\path\to\UnityProject" --dry-run
-npx --yes --package "$NccgsArchive" nccgs install --project "D:\path\to\UnityProject"
-npx --yes --package "$NccgsArchive" nccgs doctor --project "D:\path\to\UnityProject"
-```
-
-Before 2.1.0 publication, use this local archive. Published nccgs@2.0.0 does not
-include the uninstall command. Use update for an existing installation only after reviewing its
-dry run and backup. See [the tested upgrade matrix](docs/UPGRADE-QUALIFICATION.md) and its limits.
-The installer preserves customized files and unrelated Claude settings, records
-overrides and backs up replacements. Do not use --force as an ordinary update path.
-
-New installs default to standalone with `claude-standalone`. Existing installations
-without execution.json remain compatibility. To migrate explicitly:
-
-```powershell
-npx --yes --package "$NccgsArchive" nccgs migrate-execution --project "D:\path\to\UnityProject" --to standalone --model-profile claude-standalone --dry-run
-```
-
-Inspect the dry run, then remove `--dry-run`. Omitting `--model-profile` preserves
-existing model routes. See [upgrading](UPGRADING.md) for rollback and historical
-task handling. Updating files never converts an old approval into a new one.
-
-## Operating flow
-
-Start a task session with native permissions already selected:
-
-```powershell
-npx --yes --package "$NccgsArchive" nccgs run --project "D:\path\to\UnityProject"
-```
-
-New installations default to task autonomy `full`. Existing installations keep
-their setting; use `update --autonomy full` to select it explicitly. For a resumed
-session, append `-- --resume SESSION_ID`. If Claude is not on PATH, pass
-`--claude "C:\path\to\claude.exe"` before the separator. Model and tool settings
-are inherited. Ordinary Bash, report writing and task work outside the project
-need no repeated permission; explicit deny rules and integrity/destructive-action
-guards remain effective. See [task autonomy](.claude/nccgs/protocols/task-autonomy.md).
-
-```text
-User request -> bounded contract + mode + risk
-            -> reserved Claude Code worker(s)
-            -> targeted evidence + independent internal review
-            -> honest report of outcome and unresolved findings
-            -> configured external/manual/acceptance gates -> DONE
-```
-
-| Workflow | Outcome |
-|---|---|
-| `/work` | Implement a feature, fix, refactor, tool or content change |
-| `/design` | Plan design and requirements |
-| `/audit` | Examine the project without source edits |
-| `/migrate-project` | Review and apply authorized project adoption changes |
-| `/test` | Verify declared requirements |
-| `/review` | Independent review |
-| `/release` | Prepare a verified release candidate |
-| `/status` | Read state, evidence, blockers and next action |
-
-FAST uses Sonnet/medium; STANDARD uses Opus/medium; CONTROLLED uses Opus/high.
-Independent review uses Opus/high. These aliases are requested routes. Real Claude
-model resolution and provider access must be verified in the target session.
-Historical `astra-claude` routes remain available under compatibility execution.
-
-## Remove NCCGS when it is no longer needed
-
-Close Claude Code sessions, then preview and apply from an external terminal:
-
-```powershell
-npx --yes --package "$NccgsArchive" nccgs uninstall --project "D:\path\to\UnityProject" --dry-run
-npx --yes --package "$NccgsArchive" nccgs uninstall --project "D:\path\to\UnityProject"
-```
-
-This also supports an existing 2.0.0 installation without updating it first.
-Game source, customized files and `.nccgs` history/budgets remain. The command
-removes unchanged managed files and exact framework hooks/imports, saves a backup
-and refuses active or uncertain work. Use `uninstall --rollback ID` to restore
-that backup when no later edits conflict. Shared permission/model settings stay
-because their ownership was not recorded by earlier installers.
-See [details and recovery](docs/UNINSTALL.md).
-
-## Maintainer verification and limits
-
-These commands run from a source/package checkout. They are not extra steps for
-users installing with npx, and do not initiate a Claude Code game run.
-
-```text
-npm run policy:build
-npm run policy:check
-npm run validate
-npm test
-node scripts/benchmark-v15.mjs --out PATH_TO_JSON
-npm pack --dry-run
-```
-
-Generated catalogs, workflow wrappers and agents come from pipeline.json and the
-agent library. Do not patch generated model pins directly. Official Unity Skills
-remain the source for generic Editor/asset mechanics.
-
-Reservations and direct edit hooks enforce declared path ownership. Arbitrary
-implementation shell/MCP tools are not filesystem-sandboxed; review actual changes.
-Background or uncertain workers retain slots until final correlated evidence or
-external reconciliation. Worker telemetry has explicit coverage and excludes
-unobserved coordinator/human/provider activity. It is not account-wide billing.
-
-See [the Claude Code pilot and benchmark protocol](docs/NCCGS-1.5-PILOT.md),
-[the standalone workflow](.claude/nccgs/protocols/standalone.md), and
-[the 1.4.1 integrity guide](docs/NCCGS-1.4.1.md).
+# NCCGS — New Claude Code Game Studio
+
+**2.1.1** is the prepared local release for standalone game-development
+orchestration in **Claude Code**, derived from the accepted local R11 r05
+baseline. The registry already contains a different `nccgs@2.1.0` tarball. This prepared
+2.1.1 package derives from the accepted local R11 r05 qualification; it does not
+replace or retroactively qualify the registry artifact. No npm tags are changed.
+
+It retains handoff automation, bounded task slicing and Unity repair, protected
+final-review allocation, managed Unity settlement and exact first-import output
+declarations. New repair steps block unsaved loaded scenes before execution.
+See [handoff operations](docs/HANDOFF-AUTOMATION.md), [bounded repair](docs/BOUNDED-REPAIR.md),
+[task slicing](docs/TASK-SLICING.md), [uninstallation](docs/UNINSTALL.md) and [release notes](CHANGELOG.md).
+
+Claude Code handles intake, planning, specialist dispatch, testing, independent
+review and reporting. External review can be configured separately. This major
+version introduces standalone defaults and explicit execution-policy migration
+from the older compatibility workflow; existing task obligations remain preserved.
+
+## Validation and limits
+
+The frozen exact RC4 archive was exercised through actual Claude Code on a
+controlled Unity scene repair: dirty-scene refusal preserved unsaved content,
+a fresh worker repaired the target, deterministic PARTIAL handoff required no
+continuation, all five checks and 28 criteria/385 assertions passed, and a fresh
+independent reviewer returned unconditional PASS before normal task closure.
+This is scoped workflow evidence, not broad game-quality or productivity proof.
+
+Version 2.0.0 keeps those runtime guards and adds a receipt publisher so external
+Editor stop evidence is available inside the reviewer's project. It does not
+operate processes or alter acceptance. See [Editor evidence](docs/EDITOR-SESSION-EVIDENCE.md).
+The exact 2.0.0 archive subsequently passed its scoped native Unity qualification
+and was published on 2026-10-28. Frozen release receipts retain its tests and
+independent reviews. The accepted local R11 r05 baseline accounts for 353 passing
+top-level cases and 13 explicitly host-skipped link cases, with independent Fable
+static review PASS and no mandatory conditions. The 2.1.1 preparation changes
+package identity and current guidance only; runtime code, profiles, role ceilings,
+guards and budgets remain unchanged. Its exact package/release checks are recorded
+in the accompanying qualification receipts. Effective native effort stays UNKNOWN;
+this does not claim managed-worker completion or a new Claude Code/Unity run.
+Earlier negative verdicts remain unchanged in their separate receipts.
+Frozen history remains intact.
+See [release status](docs/RELEASE-STATUS.md), [upgrading](UPGRADING.md),
+[Vietnamese quick start](docs/HUONG-DAN-MIGRATE-VA-SU-DUNG.md) and [decisions](MEMORY.md).
+
+## What changed
+
+Standalone `/design`, `/design-review` and `/story-readiness` now load the
+[design lifecycle](.claude/nccgs/protocols/design-lifecycle.md). New installs require
+a reviewed design handoff before dependent feature implementation; criteria and
+their source/threshold authority are pinned across that handoff. The fixed
+`task design-validate` command supports plan tasks without granting arbitrary
+shell execution. Existing policies and frozen evaluations retain their obligations;
+see [upgrading](UPGRADING.md). Schema validation is not semantic or gameplay acceptance.
+
+
+
+NCCGS includes a [shared Ponytail package](.claude/nccgs/packages/ponytail/README.md)
+for technical planning, implementation, supplementary complexity review and known
+code limits. The coordinator and agents load it by assignment through their common
+contracts. It borrows selected guidance from DietrichGebert/Ponytail under MIT;
+the [original license](.claude/nccgs/packages/ponytail/LICENSE) and
+[attribution/change notice](.claude/nccgs/packages/ponytail/NOTICE.md) ship with every
+installation. It preserves full design and QA obligations; no separate Ponytail
+plugin or global hooks are installed. Broad comparative effectiveness remains unmeasured.
+
+- Separate execution policy, model profile and external review (`off`, `advisory`,
+  `required`); project updates preserve existing obligations.
+- Task contracts for `implement`, `audit` and `plan`, with conservative
+  FAST/STANDARD/CONTROLLED classification, reasons, inputs and acceptance gates.
+- Durable dispatch reservations, path ownership checks, two-worker ceiling,
+  independent reviewer slot, ancestor quotas and recovery without budget refunds.
+- Final evidence tied to task, source, contract, runtime invocation and raw source.
+  Missing model/effort/usage is explicit; requested settings do not prove execution.
+- Reports for success, partial, blocked and failed outcomes; strict successful
+  closure, optional external receipts and explicit acceptance when configured.
+- Execution migration with dry-run, archived configuration and guarded rollback.
+
+The studio retains 47 roles (13 active by default, including Game Designer), 36 NCCGS workflows, scoped
+rules, lifecycle hooks and unmodified official Unity Skills. Enable specialists
+when their expertise resolves a concrete problem; do not dispatch the whole studio.
+
+## Try the local 2.1.1 package
+
+Requirements: Claude Code and Node.js 18+. Unity CLI/Pipeline is needed for live
+Unity operations. Run against a disposable Unity project first:
+
+Use the locally prepared archive and its accompanying SHA-256 manifest:
+
+```powershell
+$NccgsArchive = "D:\path\to\nccgs-2.1.1.tgz"
+npx --yes --package "$NccgsArchive" nccgs install --project "D:\path\to\UnityProject" --dry-run
+npx --yes --package "$NccgsArchive" nccgs install --project "D:\path\to\UnityProject"
+npx --yes --package "$NccgsArchive" nccgs doctor --project "D:\path\to\UnityProject"
+```
+
+Before 2.1.1 publication, use this exact local archive. The registry's existing
+2.1.0 has different bytes and is not the accepted local R11 r05 archive. Published
+nccgs@2.0.0 does not include the uninstall command. Use update for an existing installation only after reviewing its
+dry run and backup. See [the tested upgrade matrix](docs/UPGRADE-QUALIFICATION.md) and its limits.
+The installer preserves customized files and unrelated Claude settings, records
+overrides and backs up replacements. Do not use --force as an ordinary update path.
+
+New installs default to standalone with `claude-standalone`. Existing installations
+without execution.json remain compatibility. To migrate explicitly:
+
+```powershell
+npx --yes --package "$NccgsArchive" nccgs migrate-execution --project "D:\path\to\UnityProject" --to standalone --model-profile claude-standalone --dry-run
+```
+
+Inspect the dry run, then remove `--dry-run`. Omitting `--model-profile` preserves
+existing model routes. See [upgrading](UPGRADING.md) for rollback and historical
+task handling. Updating files never converts an old approval into a new one.
+
+## Operating flow
+
+Start a task session with native permissions already selected:
+
+```powershell
+npx --yes --package "$NccgsArchive" nccgs run --project "D:\path\to\UnityProject"
+```
+
+New installations default to task autonomy `full`. Existing installations keep
+their setting; use `update --autonomy full` to select it explicitly. For a resumed
+session, append `-- --resume SESSION_ID`. If Claude is not on PATH, pass
+`--claude "C:\path\to\claude.exe"` before the separator. Model and tool settings
+are inherited. Ordinary Bash, report writing and task work outside the project
+need no repeated permission; explicit deny rules and integrity/destructive-action
+guards remain effective. See [task autonomy](.claude/nccgs/protocols/task-autonomy.md).
+
+```text
+User request -> bounded contract + mode + risk
+            -> reserved Claude Code worker(s)
+            -> targeted evidence + independent internal review
+            -> honest report of outcome and unresolved findings
+            -> configured external/manual/acceptance gates -> DONE
+```
+
+| Workflow | Outcome |
+|---|---|
+| `/work` | Implement a feature, fix, refactor, tool or content change |
+| `/design` | Plan design and requirements |
+| `/audit` | Examine the project without source edits |
+| `/migrate-project` | Review and apply authorized project adoption changes |
+| `/test` | Verify declared requirements |
+| `/review` | Independent review |
+| `/release` | Prepare a verified release candidate |
+| `/status` | Read state, evidence, blockers and next action |
+
+FAST uses Sonnet/high; STANDARD uses Opus/high; CONTROLLED uses Opus/xhigh.
+Leadership requests Opus/xhigh. Independent review uses Fable/high. These aliases are requested routes. Real Claude
+model resolution and provider access must be verified in the target session.
+Historical `astra-claude` routes remain available under compatibility execution.
+
+## Remove NCCGS when it is no longer needed
+
+Close Claude Code sessions, then preview and apply from an external terminal:
+
+```powershell
+npx --yes --package "$NccgsArchive" nccgs uninstall --project "D:\path\to\UnityProject" --dry-run
+npx --yes --package "$NccgsArchive" nccgs uninstall --project "D:\path\to\UnityProject"
+```
+
+This also supports an existing 2.0.0 installation without updating it first.
+Game source, customized files and `.nccgs` history/budgets remain. The command
+removes unchanged managed files and exact framework hooks/imports, saves a backup
+and refuses active or uncertain work. Use `uninstall --rollback ID` to restore
+that backup when no later edits conflict. Shared permission/model settings stay
+because their ownership was not recorded by earlier installers.
+See [details and recovery](docs/UNINSTALL.md).
+
+## Maintainer verification and limits
+
+These commands run from a source/package checkout. They are not extra steps for
+users installing with npx, and do not initiate a Claude Code game run.
+
+```text
+npm run policy:build
+npm run policy:check
+npm run validate
+npm test
+node scripts/benchmark-v15.mjs --out PATH_TO_JSON
+npm pack --dry-run
+```
+
+Generated catalogs, workflow wrappers and agents come from pipeline.json and the
+agent library. Do not patch generated model pins directly. Official Unity Skills
+remain the source for generic Editor/asset mechanics.
+
+Reservations and direct edit hooks enforce declared path ownership. Arbitrary
+implementation shell/MCP tools are not filesystem-sandboxed; review actual changes.
+Background or uncertain workers retain slots until final correlated evidence or
+external reconciliation. Worker telemetry has explicit coverage and excludes
+unobserved coordinator/human/provider activity. It is not account-wide billing.
+
+See [the Claude Code pilot and benchmark protocol](docs/NCCGS-1.5-PILOT.md),
+[the standalone workflow](.claude/nccgs/protocols/standalone.md), and
+[the 1.4.1 integrity guide](docs/NCCGS-1.4.1.md).
+
+## Direct Unity transport
+
+Unity uses CLI/Pipeline only. `nccgs run` supplies `--strict-mcp-config --mcp-config '{"mcpServers":{}}'` by default, preserving prompt/model/permission policy. An external operator may explicitly supply a JSON non-Unity MCP configuration or JSON-file path; Unity/Coplay/AnkleBreaker server identities and routes are rejected. Global MCP settings and plugins are never edited. The universal transport hook also refuses Unity MCP tools and ordinary registration/launch shell commands before task/profile routing.
+
+Generic Unity Skills retain their upstream hashes and licenses. Their optional MCP examples do not override this package's [Unity boundary](.claude/nccgs/protocols/unity-boundary.md). These checks do not sandbox arbitrary/encoded programs or a hostile client. Actual source review, native permissions and measured runtime evidence remain necessary.
+
+R07 review repair: Claude Code declares its MCP-config option variadic. The one validated config is therefore placed at argv end or immediately before the explicit option terminator, preserving positional prompts and preventing an unchecked second config. Recovery requires that canonical boundary without rewriting retained state. Ordinary Claude/Codex Unity MCP registration is denied; non-Unity community server/method names remain available. Provider instruction-pattern warnings remain fail-closed; qualification handoffs describe settings in prose instead of reproducing configuration JSON.