HackPit is plan, operate and go-deeper surfaces around one idea: the AI does the work, and you decide how much of it runs on its own. Here's everything it can do.
You approve every single command. The human is the wall — nothing runs without your click.
Passive recon and scans auto-run; every exploitation-class action is queued for you to approve or skip.
Everything fires on its own — bounded by your written Rules of Engagement, scope, a fire budget and a tamper-proof log.
Autonomy is off until you flip two switches. The repeater and anything dangerous never auto-fire, and a kill-switch halts it mid-run.
Describe a target, get an ordered recon → privesc plan, grounded in real commands.
Scoped targets, saved progress, evidence, and exam-ready reports.
Load BloodHound, see the route to Domain Admin, walk it edge by edge.
Map a cloud account's IAM to the shortest path to admin, across AWS/Azure/GCP.
Stitch web → cloud → on-prem AD into one route to the objective.
Static + AI code audit across many languages, including Web3 contracts.
Compose your own reusable prompt-step playbooks over the audit fan-out.
The approval queue — every agent-proposed command, each with a second opinion.
The guided loop — propose, approve, run, parse the result back into state.
A scoped domain → recon jobs → a ranked, ready-to-hit attack surface.
Mine hidden params and paths — each a pre-filled hand-off to the next tool.
Pull a target's JavaScript, mine endpoints, params and leaked secrets.
A real PTY in the box — vim, top, msfconsole, named parallel sessions.
A clean, per-command shell — the transcript your reports are built from.
Compose, send, replay and diff web requests. Human-only, always.
Record what your tools send, then actively scan the captured traffic.
One request, marked positions, a payload set — one approval.
Fire one request N times in the same instant to beat a check-then-act window.
Probe a front-end/back-end pair for request-smuggling desync — safe by default.
Probe a shared cache for poisoning and deception — safe detection, gated confirm.
Decode and tamper JWT, OAuth/OIDC and SAML, then replay through the repeater.
Spray captured creds, crack captured hashes — a hit lights the AD graph.
Template-scan the scoped target; matches become severity-ranked findings.
Catch and drive one live shell — no multi-implant fleet, all gated.
Pivots and DNS/TCP tunnels through the sandbox.
Drive an external Windows/AD host live over WinRM.
Generate-only payload shaping. It never auto-runs.
A canary for blind SSRF/XXE/RCE — the callback is the proof.
Where a callback lands — a local interface, or a VPS redirector.
Route outbound through a rotating pool of attributable proxy IPs.
One-command mobile bench: boot → install → cert → proxy → "log in".
Any generated command gets one honestly-labelled alternative + a which-is-better verdict.
For every action, see exactly what a defender would see — telemetry, ATT&CK, the alert that fires.
OSCP / CPTS / H1 reports, built from what you actually did, CVSS computed not asserted.
155 tools catalogued, each with a copy-ready command, target already filled in.
On a standing engagement, get alerted the moment a new asset appears.
Let another AI agent see your engagement — eyes, not hands (execution is opt-in).