| name | talos-governance-agent |
| description | Use when working on Talos governance-agent or TGA behavior, supervisor decisions, approval gates, minted capabilities, tool-call orchestration, or operator workflows. Preserve no-direct-tool-call rules, supervisor authority, schema validation, deterministic effect tracing, and audit chain reconstruction. |
Talos Governance Agent
Use this skill for governance-agent and TGA-oriented work.
Load only what the task needs:
references/tga-map.md../planning/tga-plan.md../planning/program-anchor-index.md- The closest module
AGENTS.md
Workflow:
- Identify where the change sits in the TGA flow: propose, decide, execute,
observe, or audit.
- Confirm the approval and capability model before editing. Nontrivial writes
must remain supervisor-gated.
- Preserve the execution boundary: tools are reached via Talos-protected paths,
typically through the gateway, not by direct third-party calls.
- Update schemas, validation, and idempotency handling when payload shapes or
decision logic change.
- Add or update tests that prove the chain can still be reconstructed across
action request, supervisor decision, tool call, and resulting effect.
Guardrails:
- Do not add direct tool bypasses around gateway or MCP protections.
- Do not weaken supervisor approval for write or high-risk actions.
- Do not accept unvalidated or non-deterministic model output as an execution
artifact.
Done checklist:
- TGA stage and risk tier identified.
- Supervisor and capability invariants preserved.
- Validation and audit chain coverage updated.
- Final summary calls out the affected stage and verification scope.
