mr-robot - SKILL.md Agent Skill

name: mr-robot description: >- Pressure-tests the scoped surface from an adversarial viewpoint by tracing exploit paths, abuse cases, and chaining conditions. Use when the user asks to run adversarial review, pressure-test this code, think like an attacker, or look for exploit chains — even when they only ask "how would someone break this?". Focuses on chained, attacker-driven scenarios; hands single defensive flaws with no abuse sequencing to `review/security-review` and security-governance decisions to `review/cso`. version: 1.0.0

Mr Robot

Purpose

Pressure-tests the scoped surface from an adversarial viewpoint by tracing exploit paths, abuse cases, and chaining conditions.

Use This Skill When

This lens thinks like an attacker — chaining weak checks, unsafe defaults, and race windows into a concrete abuse narrative:

"run adversarial review" / "think like an attacker" — start from attacker entry points and trust boundaries
"pressure-test this code" — probe what breaks under hostile, not merely unexpected, input
"look for exploit chains" — sequence individual weaknesses into a reachable exploit

Route elsewhere when the item is a standalone defensive flaw with no meaningful chaining (review/security-review) or a governance / accepted-risk decision (review/cso).

Inputs

Code surface under review with its attack surface, entry points, and trust-boundary topology.
Security-review findings and dependency advisories that identify candidate exploit primitives.
Operational context such as deployment model, exposed services, and external dependency trust levels.
Adversarial-review priorities such as high-value assets, attacker profiles, excluded abuse cases, or accepted-risk boundaries.

Outputs

Adversarial assessment with traced exploit paths, abuse cases, and chaining conditions.
Finding list ranking each exploit path by severity, blast radius, and the conditions required to trigger it.
Adversarial lens packet for review/code-chief with exploit chains, prerequisites, blast radius, and out-of-scope assumptions.

Workflow

Identify plausible attacker starting points, trust boundaries, and high-value state changes inside the scoped surface before proposing any exploit path.
Chain weak checks, unsafe defaults, race windows, and privilege transitions into concrete abuse narratives instead of listing isolated smells.
Distinguish confirmed exploit chains from speculative attack ideas, then record preconditions, blast radius, and containment options for every major item.
Deliver an adversarial packet to review/code-chief with exploit narratives, required hardening, and any follow-up that review/security-review should validate further.

Required Contracts

Before/After Evidence: Capture observable state before and after each intervention so improvements can be verified instead of asserted.
Shared severity: Report findings with the shared four-tier model so upstream and downstream packages interpret risk consistently.
Save-Protocol Adherence: When a Save Context block is received from the delegating orchestrator with Persistence active: yes, write deliverables to the provided save path. Saving is mandatory when persistence is active.

Collaboration Surface

review/code-chief
review/gatekeeper-code

Review Expectations

Trace every exploit path through concrete code, configuration, and dependency evidence — not through theoretical risk categories.
Distinguish confirmed exploitable chains from plausible-but-unproven attack scenarios so remediation prioritizes correctly.
Deliver findings that review/code-chief can merge into the consolidated review without re-tracing the attack surface.

Skip Rule

Skip only when the surface required by the review lens does not exist, such as a change with no attack surface or executable behavior to probe (docs-only or comment-only changes).

Failure Modes

Scenario	Response
A candidate exploit chain depends on trust boundaries, rate limits, or runtime controls that are not present in the supplied artifacts	State the missing defensive boundary explicitly and keep the chain conditional instead of pretending the missing controls do not exist.
Demonstrating the attack path would require unsafe execution or handling live secrets	Validate by reasoning through the full exploit chain and citing the exact vulnerable code, configuration, and inputs that make it reachable. Construct a minimal non-destructive proof — for example, a crafted payload shown inline as a code block — to make the chain concrete without executing it against live systems. Never run live exploits, exfiltrate real data, or handle real secrets to prove a finding. Describe the risk and recommend a contained validation environment (e.g., a sandboxed staging replica) for any step that cannot be safely demonstrated through reasoning alone.
Several weak links exist but the exact chaining order is uncertain	Break the chain into validated and unvalidated segments so the report stays honest about what is proven.
The issue reduces to a standard defensive security flaw with no meaningful abuse sequencing	Hand the item back to `review/security-review` and keep the adversarial packet focused on chained or attacker-driven scenarios.
The threat model or attacker profile is undefined or absent	State explicit assumptions — default to a standard external-attacker model (unauthenticated, network-accessible, motivated by data exfiltration or service disruption) — note the assumed scope, and flag that the findings may undercount risk if the real attacker profile is more privileged.

Save Protocol

When a ### Save Context block is included in the delegation prompt with Persistence active: yes:

Write deliverables (reports, evidence bundles, review packets) to the save path specified in the Save Context block.
Use filenames that match the deliverable type, such as deliverable_{name}.md, report_{name}.md, or review-packet.md.
Do not create or manage _phase-state.md — that is the delegating orchestrator's responsibility.

When Save Context is absent or Persistence active: no, skip all save operations and deliver output inline as usual.

References

references/workflow.md for the detailed operating sequence and decision rules.
references/examples.md for concrete request patterns and response shapes.

Packaging Notes

Package SKILL.md, references/workflow.md, and references/examples.md together. Keep generated reports and archives outside the skill directory.