Architecture Pattern — Subagents
Pattern: Architecture
Component: subagents.md
Version: 1.1 | Updated: 2026-07-16
| Subagent | Role | When invoked | Output |
|---|---|---|---|
| Alternatives Generator | Forces ≥2 viable designs | After drivers clear | Options table |
| Failure Walker | Dependency-down and partial-failure scenarios | Before review | Scenario → mitigation table |
| Threat Boundary Mapper | Identity, data, tenancy, tool edges | Before review | Trust-boundary notes |
Alternatives Generator contract
- Options must be implementable under stated constraints
- No “do nothing” strawman unless it is genuinely viable
- Each option names primary trade relative to QA scenarios
Failure Walker contract
- Cover: dependency unavailable, timeout/retry storm, poison message, clock skew, region loss (if multi-region)
- Each scenario needs detection signal + mitigation
Threat Boundary Mapper contract
- Mark where untrusted input enters
- Mark tenancy enforcement points
- Mark AI tool allowlists if tools exist
- Escalate unresolved Critical boundaries — do not approve around them
Version: AIES v1.0.0✏️ Edit this page on GitHub