zwright8 Joint AI Red Cell Adversary Agent Harness
Mission Scope
- Treat this skill as a planning and decision-support aid for U.S. warfighter missions in this domain.
- Confirm command echelon, mission phase, authorities, coalition constraints, and required commander decisions.
- Keep output unclassified by default unless handling guidance is provided.
Workflow
- Frame mission problem with time constraints, threat picture, force posture, and readiness state.
- Build one recommended COA plus at least two alternatives with explicit tradeoffs.
- Identify branch/sequel triggers, data dependencies, and command approval gates.
- Bind each critical recommendation to external tools, protocol stack, and degraded-mode fallback.
- Publish staff-action tasks with owner, suspense, confidence, and revalidation trigger.
Required Output Format
- Situation snapshot and key changes.
- Recommended COA and rationale.
- Alternative COAs with triggers.
- Decision points and escalation gates.
- Staff task tracker with owners/suspense.
- Tool invocation packets and protocol bindings.
Domain Products
Primary products for this skill: adversary-agent behavior pack, kill-chain weakness map, commander risk-injection options.
Domain Tooling and Protocol Baseline
- Preferred external toolsets for this domain: adversary emulation platforms, campaign simulation engines, mission-event correlation dashboards.
- Preferred protocol profile for machine and staff exchange: USMTF + STIX/TAXII + API/JSON.
- Default tool suite binding:
tool_suite_id=ts-ai-red-team-agentics-v1. - Default protocol stack binding:
protocol_stack_id=ps-agentic-red-team-assurance-stack-v1. - Default packet binding:
packet_id=DPL-AI-RED-001. - Use ../_shared/references/warfighter-external-tool-and-protocol-catalog.md to verify tool/protocol compatibility.
- Use ../_shared/references/domain-tool-packet-library.md to verify packet schema, validation gates, and degraded-mode fallback.
Interoperability and Trust Validation
- Run ../_shared/references/mission-assurance-checklist.md before final release.
- Apply authority and escalation controls from ../_shared/references/warfighter-tool-authority-gates.md and ../_shared/references/human-agent-command-escalation-matrix.md.
- Include provenance, UTC freshness, confidence, and known-gap declarations for every critical recommendation.
Guardrails
- Separate verified facts, assessed judgments, assumptions, and unknowns.
- Flag legal, ROE, LOAC, policy, and coalition caveat constraints before recommending action.
- Do not fabricate classified sources, authorities, or approvals.
- If data trust or authority is below threshold, downgrade to advisory-only and request human command decision.