chore(gds): resync projections

.claude/CLAUDE.md

@@ -1 +1,67 @@
-@../AGENTS.md
+<!--
+GENERATED FILE - DO NOT EDIT DIRECTLY
+generator: gds
+bundle: 0.1.0-dev
+source-commit: 97e8bbaa3a0734b156b03bad704503bc46d7575b
+input-digest: sha256:3f9e642667c8c50ea2fc6b70f76ddff998c0691aeb427bcbbb314d0d425f0d36
+output-digest: sha256:a6727bddefbe66b0ae07503bb6b76535291812139d5972a14c4edaeb67c62e8a
+edit-source:
+  - .gds/repository.yaml
+  - policies/base/repository-default.yaml
+  - policies/owners/organization-default.yaml
+  - policies/roles/public-module.yaml
+  - templates/agents/repository.md.tmpl
+  - templates/github-actions/go.yml.tmpl
+  - templates/harnesses/claude.md.tmpl
+-->
+# Claude Code repository contract
+
+## Scope
+
+- GDS repository ID: `repo_01KYFD7GYZR9531WFQWV0KABZF`.
+- Roles: `module`.
+- Canonical repository facts: `.gds/repository.yaml`.
+- Applied policy bundle: `.gds/bundle.lock.yaml` (`0.1.0-dev`).
+- This is a first-class Claude Code projection compiled from the same typed
+  inputs as `AGENTS.md`; neither projection is a manual policy source.
+
+## Repository boundaries
+
+- Treat this Git repository as one independent mutation boundary.
+- Preserve unrelated dirty changes, branches, worktrees, and submodules.
+- Run `gds context --json` before work crosses repository boundaries.
+- Do not edit generated projections; change the declared canonical input and
+  regenerate.
+
+## Safety
+
+- External writes require explicit approval: `true`.
+- Generated projection edits: `forbidden`.
+- Private parent context persistence: `forbidden`.
+- Visibility: `public`; data: `public`.
+
+## Verification commands
+
+- Test: `python3 cli-tools/validate_public_contracts.py`.
+
+## Claude workflow routing
+
+- Start here: run `gds-orient` (or `gds context --json`) to resolve scope before
+  any cross-repository work.
+- Active skill profiles: `core, module`. Five profiles exist in total
+  (`core`, `estate-admin`, `module`, `device`, `portfolio`); only the listed ones
+  are active for this repository. The catalog is `skills/registry.yaml`, and each
+  skill lives under `skills/canonical/<name>/SKILL.md`.
+- Load procedural detail from the applicable installed GDS skill projection or
+  plugin only when the task matches it.
+- Destructive workflows remain explicit-only and still require their concrete
+  plan and approval gates.
+- Treat documentation and Serena memories as derived evidence, never mutation
+  authority.
+
+## Done
+
+- Required checks pass or are explicitly reported `NOT_PROVEN`.
+- Every affected Git boundary and remote result is classified.
+- No secret, private-context leak, unrelated change, or unapproved projection
+  drift is introduced.

.gds/bundle.lock.yaml

@@ -0,0 +1,20 @@
+# GENERATED FILE - DO NOT EDIT DIRECTLY
+schema_version: 1
+
+bundle:
+  version: "0.1.0-dev"
+  release_sequence: 0
+  channel: "development"
+  source_commit: "97e8bbaa3a0734b156b03bad704503bc46d7575b"
+  digest: "sha256:fd66beb15c847e94b68a00083df5f218c22f4e1174d39d4bd10ed7fb321b13bc"
+
+projection:
+  input_digest: "sha256:3f9e642667c8c50ea2fc6b70f76ddff998c0691aeb427bcbbb314d0d425f0d36"
+  output_digest: "sha256:4fa2ada815ed3c958865e68a364a97c60e07f830910fb5301a760466d5ab134b"
+  files:
+    - path: ".claude/CLAUDE.md"
+      digest: "sha256:e27070b595af4c33935241494cd004f81edf4b31633bde94430ec22cea9d4fbe"
+    - path: ".gds/compiled-policy.json"
+      digest: "sha256:c73ba67ecbf7671b853b72eff4d281df94e5f0d85c4fe4397abd608219b062be"
+    - path: "AGENTS.md"
+      digest: "sha256:6170a682da481be2d705a24af4b038c65724d4a2dbeb54db375c2309498346d1"