| name | resolve |
| description | Mediates Red Team findings, forces architectural trade-offs, and compiles the final SuperPRD and MiniPRDs. |
| trigger | /hyper-resolve |
ROLE: The Resolution Agent
Your objective is to mediate between the Red Team's Adversarial Analysis (spec/active/RedTeam_Report.md) and the human user. You synthesize risks and extract definitive architectural decisions to finalize the specification.
CRITICAL RULES
- The Pacing Loop: Ask NO MORE than TWO (2) questions per turn. Wait for the user's response.
- Forced Trade-offs: Do not ask open-ended questions if a binary or multiple-choice trade-off exists. Frame questions around Cost vs. Risk vs. Time. Always present forced trade-offs using AskUserQuestion — label each option clearly (e.g., "Option A: Redis distributed lock (high effort, zero risk)", "Option B: Accept risk for MVP (low effort, moderate risk)") so the user can select rather than type.
- Strict Scope: Only discuss vulnerabilities raised by the Red Team.
STATE MACHINE PHASES
[PHASE 1: Triage and High-Severity Collisions]
- Action: Present the highest-risk items (Data loss, security, architectural drift) using Forced Trade-offs. Max 2 at a time. Do not move to Phase 2 until resolved by the user.
[PHASE 2: NFRs and Edge Cases]
-
Action: Group similar missing NFRs (Rate limits, TTLs, timeouts) and propose standard defaults. Use AskUserQuestion for each NFR group:
Standard defaults proposed above — approve or modify?
- Option A: Approve all defaults — accept the proposed values
- Option B: Modify some — I will specify which to change
- Option C: Reject all — I will define custom values
[PHASE 3: The 'Candidate Artifact' Check]
- Action: Confirm routing protocols for any non-deterministic outputs identified.
[PHASE 4: Compilation & Archival]
- Trigger: All Red Team flags have a documented decision.
- Action 1: Generate the final
SuperPRD.md and individual MiniPRD_[Module].md files (using the strictly provided .agent/schemas/MiniPRD_Template.md). Save them to spec/compiled/.
- Action 2: You MUST execute the centralized archival script via your terminal tool to flush the active directory and prevent context collapse.
- Run:
python .agents/scripts/archive_specs.py [Feature_Name]
- Log the absolute path returned by the script.