Deliver the autonomous full-pentest command

[maajix]

Parent

#5

What to build

Add a full-pentest command that autonomously runs recursive recon and pentest as a bidirectional loop, performs bounded bypass escalation, kill-chain review and convergence challenges, then generates the report with an accurate terminal outcome.

Acceptance criteria

• One full-pentest invocation runs the engagement through recon, testing, authorized exploitation, challenge, and reporting without manual command chaining.
• A pentest discovery can reopen recon and later return to testing automatically.
• Recon, pentest, and report remain independently usable with their documented focused semantics.
• The command resumes correctly from every durable campaign phase after interruption.
• Reporting starts only when coverage, work, leases, bypass review, chain review, and convergence challenge are terminal for the current material digest.
• The report distinguishes converged, incomplete, operator-blocked, and budget-exhausted outcomes and includes remaining coverage where applicable.
• Existing scope, evidence, audit, rate, redaction, and rules-of-engagement behavior remains enforced end to end.

Blocked by

• Use one engagement identity across command phases #7 — Use one engagement identity across command phases
• Guarantee fair recon coverage across assets #10 — Guarantee fair recon coverage across assets
• Escalate resistant hypotheses to a bypass specialist #11 — Escalate resistant hypotheses to a bypass specialist
• Re-evaluate kill chains after every material event #12 — Re-evaluate kill chains after every material event
• Challenge apparent convergence with isolated lead explorers #13 — Challenge apparent convergence with isolated lead explorers

[maajix]

Implemented in 417ad6f on branch agent/autonomous-campaign-coordinator. Full Python (159) + shell regression suites green.