feat: agent tool aliasing — expose /agents/run and /agents/bootstrap as MCP tools

[stackbilt-admin]

Context

The agents vertical shipped in tarotscript (#125-130). Two new worker endpoints are live:

• POST /agents/run — C-Level agent reasoning (role, intention, context, pain signals → structured analysis with dignity tensions and confidence scoring)
• POST /agents/bootstrap — domain deck research (role, domain, internal signals → bootstrap methodology + card candidates for new agent decks)

The agent-response contract is signed at @stackbilt/contracts/agent-response and defines all request/response schemas.

What's needed

MCP tool definitions wrapping the agent endpoints

agent_consult

• Wraps POST /agents/run
• Input: role (string — e.g. "cto"), intention (string), context (object — org size, stage, constraints), painSignals (string[] — raw user frustrations)
• Output: AgentRunResult — stance, domain lens, constraint, action, confidence, dignity tensions, oracle synthesis

agent_bootstrap

• Wraps POST /agents/bootstrap
• Input: role (string), domain (string — the knowledge domain to research), internalSignals (string[] — what the org already knows)
• Output: bootstrap methodology + card candidates (category breakdown, trait suggestions, governance guardrails)

Schema sourcing

All request and response types are defined in @stackbilt/contracts/agent-response. The MCP tool input/output schemas should be derived from these contracts to stay in sync — no hand-rolled duplicates.

References

• tarotscript commits: 257a6be (agents vertical), 952239e (deck-bootstrap-cast)
• tarotscript issues: #125-130