agentprivacy-entry-kit-coherence
Treat mages_city/site/skill.md as the canonical detailed entry. Labs site/skill.md, discovery.md and llms.txt and landing public/skill.md and llms.txt should point to it, not maintain contradictory full copies. Preserve each site's own purpose.
The capability source is agentprivacy-mcp/lib/experience-overview.json; the City site/experience-overview.json must match it. Verify every listed tool against the actual MCP tools/list response. Distinguish implemented local tools, observed portal state, pending live adapters and design proposals. A running farm row is not verified extension control; passing fixtures are not deployment.
Check the whole path: learning, existing key, selected disclosure, invitation approval/signing, MyTerms/task/VRC verification, exact scoped action, real receipt and private journey fold. Keep public graph edges distinct from private perspective and current access.
Verify local links, inventory equality, relevant MCP tests, City static-front checks and TypeScript for landing changes. Preserve operational records. Prepare sync from the canonical unversioned skills tree; inspect consumers still reading legacy agentprivacy-skills-v5 before using their sync scripts. Do not claim live publication until actual deployment is verified.
Related domain skill: existing foundation. This workflow adds no personas or canonical vertex assignments.
City and Star operating practice
Reconcile canonical source paths and hashes with the Guideโs forkable pages, assets, search index, sync manifest and Spellweb Guide bridge. Preserve wiki journal contributions; stop on conflicting edits to a generated item. List unresolved mappings instead of inventing vertices. On source changes, semantically review prior findings and retain corrections: refreshing a hash alone is not a new verification. Keep fixture results, source observations, hypotheses and deployed behavior distinct.