offensive-claude

What it is

offensive-claude is a spec-driven offensive-security framework for Claude Code. Beyond what the catalog summary describes, the page's actionable content is the install surface and its always-on behavior: installing as a plugin registers a SessionStart hook that injects the using-offensive-claude dispatcher into every conversation, so the scope → finding → OPSEC skill-invocation discipline is active from the first message of every session — not just when you explicitly invoke it.

How it works

Phase transitions are gated. Each gate validates that required template artifacts exist and that findings carry mandatory fields — CWE, CVSS, evidence, and ATT&CK ID — before a PASS suggests the next phase; a FAIL lists missing items. Safety is enforced in shared code under coding-mastery/scripts/_lib/: scope_guard, action_guard, http_creds, redact_headers. These "safety-critical scripts" have a pytest suite (tests/) and CI (.github/workflows/tests.yml). The .devcontainer ships a binary-analysis toolchain (rr/gdb/gcov/afl++) for the crash→exploitability pipeline, running with scoped SYS_PTRACE/SYS_PERFMON capabilities rather than --privileged.

What to do

Preferred install (registers the SessionStart hook):

/plugin marketplace add hypnguyen1209/offensive-claude
/plugin install offensive-claude@offensive-claude-marketplace

Alternative one-liner:

curl -sL https://raw.githubusercontent.com/hypnguyen1209/offensive-claude/main/install.sh | bash

Or clone and run bash install.sh, or manually copy skills/, agents/, templates/, workflows/, commands/, presets/, and CLAUDE.md into ~/.claude/. Engagement entry points are the /engage.* commands (init, scope, recon, weaponize, exploit, report, plus memory and pickup). Machine-readable ROE lives in templates/scope/scope.schema.json. Before use, note TERMS.md (acceptable-use / authorization requirement) and SECURITY.md (coordinated disclosure).

Caveats

• The recommended plugin install makes the dispatcher global to every Claude Code conversation, not scoped to engagement sessions — relevant if you use the same Claude Code install for unrelated work.
• The one-liner is a curl … | bash pattern; review install.sh before piping to a shell.
• The framework enforces process discipline (authorization, proof-before-CONFIRMED, OPSEC), but these are convention/skill guardrails, not hard technical controls — the tooling still executes real offensive actions and is gated on the operator honoring TERMS.md.