Add safe managed skill updates

[dd3ok]

Summary

• add guarded managed-skill updates with tree/revision compare-and-swap, stable skill identity, complete immutable remote provenance, and verified transactional replacement
• recover interrupted updates deterministically, preserve drifted user content, and limit destructive cleanup to exact journal-owned paths
• preserve deployments during updates and report outdated copies instead of rewriting host state automatically
• intentionally remove legacy lifecycle/provenance compatibility: unsupported archive journals and partial provenance fail closed without migration or inference
• simplify the explicit-only skill guide, technical documentation, and CI so universal checks cover behavior/build/distribution while benchmarks and package inspection remain change-specific evidence

Review results

• final reviewed range: d4832ab62fd2c1bfd7fbf02c0ab845815375c0bd...0af5b005de0bf590a17949110df8d5cf70294416
• standards and specification reviews found no remaining actionable P0-P3 code findings
• GPT Pro final review confirmed the final cleanup, found no new P0-P3 findings, and concluded PASS
• the prior schema-1 compatibility finding was closed as an approved clean break rather than addressed with migration code

Validation

• npm run test:all — 74 source tests: 73 passed and 1 expected Windows fixture skip; distribution tests 8/8 passed
• npm run pack:check
• git diff --check
• generated-artifact validation passed as part of test:all
• authoring validators passed for skills/stash, the root plugin, and adapters/codex
• GitHub CI passed on final HEAD 0af5b00: Ubuntu Node 20/24, Windows Node 24, and macOS Node 24
• npm run bench was not run on the final cleanup because routing, scoring, and golden behavior did not change; it remains available as targeted routing evidence
• local platform: Windows, Node 24.18.0
• vendor binaries were not invoked; no live vendor support claim changed

Security and compatibility impact

• stored remote provenance must be fully canonical or entirely absent; mutable revisions and partial records are rejected
• metadata-only and changed-tree updates recheck the full record and managed tree at the commit boundary
• archive schema-1 recovery is intentionally unsupported; unknown journals remain untouched and fail closed
• recovery is process-crash resilient but does not claim fsync-backed power-loss durability
• bulk managed updates commit per skill rather than as one atomic batch
• tracked deployments remain unchanged until an explicit deactivate/activate cycle
• no remote calls, telemetry, plugin mutation, or vendor setting mutation were added

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 77bfc4d611

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".