Sync Agent Workbench

Sync Agent Workbench

Workflow

1. Read AI_AGENT_GUIDE.md and AI_AGENT_PROJECT.md if present.
2. Choose the requested workflow:
   • Sync: follow .agents/prompts/sync-agent-workbench.md when present, otherwise use the upstream prompts/sync-agent-workbench.md.
   • Audit: follow .agents/prompts/audit-agent-workbench.md when present.
   • Repair: follow .agents/prompts/repair-agent-workbench.md when present.
3. Modify only the files allowed by the selected prompt.
4. Keep AI_AGENT_PROJECT.md manual after initial creation.
5. Keep vendor entrypoints thin and canonical content under AI_AGENT_GUIDE.md, .agents/prompts/, and .agents/skills/.
6. Treat install as first sync: .agent-workbench.yaml is human-owned desired config and .agent-workbench.lock.json is the agent-owned provenance/baseline ledger.
7. Use lockfile + current desired set to classify removal drift as confirmed upstream removal, confirmed removal with local edits, suspected legacy removal, deselected by local config, source changed / migration required, or local unmanaged. Never delete without explicit user confirmation; record kept removals in retainedRemovals.
8. Verify with git status --short, git diff --stat, and manifest path checks when working in the workbench repository.

Guardrails

• Do not install marketplace plugins, extensions, global/user-scope settings, or git submodules.
• Do not modify application source code during sync unless the user separately requested application changes.
• Do not create AGENT.md; use AGENTS.md.
• Do not delete downstream artifacts silently; deletion candidates must be normalized, allowlisted workspace-overlay paths and user-confirmed.

Source: KiringYJ/agent-workbench — distributed by TomeVault.