B-NXEN-agent-personas-verification-skill-static-judgement-adopted
Status: promoted/milestone
Introduce a set of agent personas and a reusable verification skill, then wire them into the task-work flow.
Verification skill
Section titled “Verification skill”A skill that performs combined verification of work:
- Static analysis — deterministic/mechanical checks.
- Judgement analysis — LLM-judgement evaluation against the intersecting set of relevant principles (docs/planning/principles/) and standards (docs/planning/standards/) for the work under review.
The skill resolves which principles/standards apply to the changed surface and scores the work against that intersecting context, surfacing violations from both the static and judgement passes.
Agent personas
Section titled “Agent personas”- Implementor — does the actual work for a task.
- Orchestrator — coordinates implementor + verification, drives the loop.
- (Plus whatever persona the verification step embodies.)
Adoption in task-work + rework loops
Section titled “Adoption in task-work + rework loops”Adopt the verification skill inside /sdlc:task-work so that after implementation, work is verified; on failure the orchestrator dispatches a rework loop back to the implementor with the verification findings, iterating until verification passes (or a bound is hit).
Open questions
Section titled “Open questions”- Where personas live (prompt fragments vs agent definitions vs skill front-matter).
- How the intersecting principle/standards context is computed (path globs in applies_to vs semantic match).
- Rework-loop bound / escalation when verification keeps failing.
- Relationship to existing /sdlc:prd-verification and /sdlc:principle-review skills.