29 lines
1.2 KiB
Markdown
29 lines
1.2 KiB
Markdown
# Config Generation Workflow
|
|
|
|
## Goal
|
|
Keep agent policy consistent and avoid drift between registry and runtime router config.
|
|
|
|
## Current Reality
|
|
- `tools/agents generate` updates:
|
|
- `config/router_agents.json`
|
|
- `config/crewai_agents.json`
|
|
- `gateway-bot/agent_registry.json`
|
|
- `services/router/router-config.yml` is still runtime-critical and currently not fully regenerated from registry.
|
|
|
|
## Update Procedure
|
|
1. Edit canonical policy in `config/agent_registry.yml`.
|
|
2. Run `python3 tools/agents generate`.
|
|
3. Manually sync runtime-critical rules in `services/router/router-config.yml`:
|
|
- llm profiles (`cloud_deepseek`, `cloud_mistral`, `cloud_grok`, local profiles)
|
|
- per-agent routing rules (`*_agent` rules)
|
|
- infra agents local-first constraints
|
|
4. Deploy router and run smoke:
|
|
- infer: all top-level agents + `monitor`
|
|
- webhook: `/telegram/webhook` + selected agent webhook
|
|
- canary: `ops/monitor_canary_summary.sh`
|
|
5. Commit policy docs + config updates in one PR/commit set.
|
|
|
|
## Hard Rules
|
|
- No temporary code-level model policy hacks in `services/router/main.py`.
|
|
- If emergency hotfix is needed, remove it in the next stabilization commit.
|