dayan-agent-factorylisted

# Dayan Agent Factory Use this Skill only after an agent design has a stable objective, role boundary, tool policy, output contract, and evaluation plan. ## Package Create one directory containing: - `AGENT.md`: role, objective, inputs, output, exclusions, and stop conditions; - `tool-policy.json`: allowlisted capabilities and forbidden external effects; - `output-schema.json`: machine-readable result contract; - `examples/accepted.json` and `examples/rejected.json`; - `EVALUATION.md`: tests, human takeover, cost/latency limits, and unresolved risks; - `README.md`: install, run, verify, and remove instructions. Reject packages with broad filesystem access, hidden state ownership, unbounded retries, self-approval of consequential actions, or no rejected fixture. Generate the package; do not install or deploy it without separate authorization.