Orchestrate multi-step, multi-agent workflows with HITL gates.
"fix the issue and verify it works", "scaffold an agent and configure knowledge").
Decomposes multi-step requests into sequenced tasks that are delegated to specialist agents (Author, Manage, Test, Troubleshoot) with HITL gates between each step.
circe-memory.md (if it exists) — understand current project state..github/circe-conventions.md — load mandatory rules.Match the user's request to a predefined workflow:
| Pattern | Workflow |
|---|---|
| "create X and deploy" / "add X and push" | Create & Deploy |
| "fix X and test" / "debug X and verify" | Fix & Verify |
| "create X, deploy, and test" / "full cycle" | Full Cycle |
| "scaffold agent and configure" / "new agent from scratch" | Scaffold & Configure |
| "create a BC agent" / "collections agent" / "sales agent" / "support agent" | BC Agent Setup — read /copilot-studio:bc-agent-blueprints for the matching blueprint |
| None of the above | Custom Workflow — analyze and propose |
Before executing anything, present the plan with numbered steps, agent assignments, and HITL gates. Wait for user confirmation.
For each step:
If a step fails:
After the final step:
/copilot-studio:update-memory with the workflow resultsWhen delegating to a specialist agent, provide:
Task: [specific action to perform]
Context: [user's original request and intent]
Project state: [from circe-memory.md — schema name, GenerativeActionsEnabled, etc.]
Previous steps: [what was already done in this workflow]
Decisions: [relevant DRs from this session]
── CIRCE EVIDENCE ──────────────────
Skills loaded: copilot-studio-conductor
Workflow executed: [workflow name] ([N] steps)
Agents delegated: [list]
HITL gates passed: [N/total]
Decision Records created: [list or "none"]
Memory updated: Sí/No
────────────────────────────────────