Bug: No centralized agent inventory — orchestrator has no source of truth for active agents
Problem:
When running multiple agents (gateway orchestrator + Claude Code + sub-agents + external models), there is no built-in mechanism to track which agents exist, their status, capabilities, or trust level. The orchestrator must manually read scattered config files and memory to reconstruct the agent landscape.
Impact:
Onboarding a new agent requires reading 20+ files manually. No single command or file answers "who is in my system right now?" Agent status lives in individual files with no aggregation. Cross-agent coordination relies on human memory, not system state. Scaling from 2 to 5+ agents becomes unmanageable.
Observed: 6 days of governance work produced 50+ memory files but no centralized agent registry. When asked "are we ready to onboard 4 new agents?" the answer required reading 8 separate documents.
Expected: A structured agent catalog (JSON or equivalent) that the orchestrator can query — listing all agents, their platform, role, status, trust level, capabilities, and dependencies. Ideally maintained automatically via hooks or heartbeat.
Environment: OpenClaw 2026.1.30, Windows 10, multi-agent setup (gateway + Claude Code + planned: Kimi, secondary gateway, secondary Claude Code)
Workaround: Manual curation of agent list in AGENTS.md markdown table. Does not scale, goes stale immediately.
Bug: No centralized agent inventory — orchestrator has no source of truth for active agents
Problem:
When running multiple agents (gateway orchestrator + Claude Code + sub-agents + external models), there is no built-in mechanism to track which agents exist, their status, capabilities, or trust level. The orchestrator must manually read scattered config files and memory to reconstruct the agent landscape.
Impact:
Onboarding a new agent requires reading 20+ files manually. No single command or file answers "who is in my system right now?" Agent status lives in individual files with no aggregation. Cross-agent coordination relies on human memory, not system state. Scaling from 2 to 5+ agents becomes unmanageable.
Observed: 6 days of governance work produced 50+ memory files but no centralized agent registry. When asked "are we ready to onboard 4 new agents?" the answer required reading 8 separate documents.
Expected: A structured agent catalog (JSON or equivalent) that the orchestrator can query — listing all agents, their platform, role, status, trust level, capabilities, and dependencies. Ideally maintained automatically via hooks or heartbeat.
Environment: OpenClaw 2026.1.30, Windows 10, multi-agent setup (gateway + Claude Code + planned: Kimi, secondary gateway, secondary Claude Code)
Workaround: Manual curation of agent list in AGENTS.md markdown table. Does not scale, goes stale immediately.