gungnir

gungnir

!checklist init ${CLAUDE_SKILL_DIR} --force

The single most valuable security mindset is to think like an attacker and actively try to break your own system — because a defense is only ever proven by a real attack, never by a clean scan or a green test. This skill is the spear: authorized, adversarial penetration testing of a system to prove (or disprove) that the defenses the aegis skill built actually hold. It drives a real attack through its whole arc — scope, recon, scan, exploit, chain, report — across six gated stages, and it will not advance past a GATE until the checklist tool clears it.

The one hard boundary, gated first and never crossed: you attack only a system you own or have explicit, written authorization to test. Attacking anything else — a third-party service, a system you don't own, infrastructure outside the agreed scope — is out of bounds, full stop, regardless of intent. This skill is for authorized, defensive validation: your own staging environment, a system you've been contracted to test in writing, or a deliberately-vulnerable practice lab. STAGE 0 makes that authorization a hard gate, and it is the one gate that is never a judgment call.

Adversarial validation is where the agent era cuts both ways, and both edges matter:

• The agent can now actually run the attack — nmap, sqlmap, an intercepting proxy — which makes continuous self-attack cheap and real. That capability is exactly why the authorization gate must be absolute: a tool that can find a hole can point at the wrong target, and the agent feels no instinct about whose system it is.
• The agent shares the blind spots of whoever built the system. If the design, the code, and now the attack are all done by agents, they share a training distribution and an imagination — so an agent attacking a system it (or a sibling) built will miss the same things it missed building it. Agreement is not evidence. Real assurance needs independent angles, and on the highest-stakes systems, a genuinely independent (human) tester.
• The agent runs a scanner, sees an empty result, and declares "secure." A clean scan means known patterns weren't found — it is necessary and nowhere near sufficient. The creative work — chaining small flaws into a real exploit, abusing business logic no scanner understands — is exactly what the agent is weakest at and what this skill exists to force.

So the rule that governs this skill: a defense is proven only by a real, authorized attack — a clean scan is not proof — and you attack only what you are permitted to. The goal is not to "pass"; it is to find the holes while they are still cheap to fix.

Discipline: finish every GATE before the next stage. GATEs are hard — never skip, batch past, or self-certify a stage you have not done. The checklist tool enforces the order; let it. Commands address stages by name.

Read references/agent-era-shifts.md first — it is the heart: what adversarial testing becomes when the attacker is an agent that shares the defender's blind spots, can run the tools, and will mistake a clean scan for a secure system. If $ARGUMENTS is trivial and has nothing worth attacking, this machinery is overkill — say so.

Speak the user's language, and confirm the authorization out loud. This skill needs the user to set the scope, confirm ownership/authorization, and own the disposition of every finding. Read their fluency and gloss a term on first use (penetration test vs scan, recon, IDOR, injection, privilege escalation, scope, rules of engagement). A user who cannot confirm, plainly, that this target is theirs to attack has not authorized it — and STAGE 0 does not clear.