Agent
Manual procedures to verify each subagent in plugins/pentest-core/agents/ works end-to-end after merge. Run once after Plan C lands. Re-run when an agent definition changes materially.
Agent
Researches a single discovered service (host:port:product:version) using public threat-intel sources. Produces a per-service markdown research file. NO Bash. NO Skill. Restricted to public threat-intel sources — must NEVER WebFetch the engagement target.
Agent
Judges the output of a pentest phase batch and returns a verdict (proceed | replan | halt). Read-only — must not dispatch other subagents, shell out, or modify engagement state. Returns supervisor-verdict JSON.
Agent
Executes a single pentest phase against assigned targets, returning structured worker-result JSON. Dispatched by the orchestration conductor (/pentest-engage). Tools restricted to Skill, Bash, Read, Write — must invoke the matching pentest- phase skill rather than raw tools.