Model hint: Use
model: "opus"when spawning subagents for this skill (deep reasoning task: critical analysis + neutralization experiments).
Inputs to read (project-local):
- •.claude/memory/project_brief.md
- •.claude/memory/repo_policy.md
Procedure:
- •Collect evidence pointers (read-only):
- •recent commits (git log)
- •key result files (list a few)
- •any draft text/figures if present
- •Call redteam-project agent.
- •Produce a report at results/redteam/YYYY-MM-DD.md:
- •Top 5 criticisms
- •fatal vs addressable
- •1 concrete neutralization experiment/analysis each
- •level of effort: 2h / 2d / 2w
Rules:
- •Only use provided sources / available repo artifacts.
- •Be harsh, technical, precise.