zero-lab skills

Portable, evidence-bound security skills for authorized defensive assessment,
penetration testing, red teaming, detection engineering, purple teaming, and
engagement operations.

The collection contains 56 skills in 10 installable plugins from
zero-lab.ai. Install a coherent bundle, or copy an
individual skill folder into any agent harness that reads SKILL.md.

[!IMPORTANT]
These skills support authorized security work. They do not grant permission,
expand scope, or replace an organization's rules of engagement. Skills that
touch adversary behavior are designed to measure exposure, preserve evidence,
and hand results back to defenders.

Install
Bundles
Skill index
Compatibility
Trust and safety
Repository structure
Contributing
Contributors
Security
License

Install

Claude Code

/plugin marketplace add zero-lab-ai/skills
/plugin menu

Select the bundle you need from the marketplace menu.

Codex

Codex reads the same marketplace metadata directly:

codex plugin marketplace add zero-lab-ai/skills
codex plugin list
codex plugin add <bundle-name>@zero-lab

ChatGPT Workspace Marketplace

Use this repository's .claude-plugin/marketplace.json for workspace imports.
Each skill includes agents/openai.yaml with its display name, description,
Zero Lab mark, and brand color.

Other agent harnesses

Install the cross-agent skills tree:

npx skills add zero-lab-ai/skills

You can also copy a single
plugins/<bundle>/skills/<skill>/ directory into the location your harness
scans. Nothing outside that skill directory is required at runtime.

Bundles

Plugin	Skills	Purpose
`agent-security`	4	Agent surfaces, prompt injection, runtime validation, and trust boundaries
`recon-operations`	9	Network, public-exposure, human, physical, and AI-system reconnaissance
`identity-security`	8	Directory, credential, privilege, and lateral-movement security
`application-endpoint-security`	6	Cloud, web, mobile, firmware, endpoint, and password assessment coverage
`network-protocol-security`	5	Wireless, traffic, protocol, command-channel, and egress security
`binary-forensics`	4	Binary-fault, memory, and rootkit triage
`detection-validation`	6	Detection engineering, validation, telemetry, resilience, and control evidence
`engagement-operations`	7	Planning, AI threat modeling, evidence proof, findings, and report assembly
`rag-security`	3	Retrieval-surface, integrity, tenant, data-exposure, and cleanup assessment
`ml-system-security`	4	Model artifacts, decision boundaries, datasets, and serving exposure

Skill index

Agent and AI systems

Install agent-security.

Skill	Use it when…
`agent-surface-audit`	Reviewing agent tools for poisoning, excess privilege, unsafe parameters, and deployment blast radius
`agent-runtime-validation`	Validating deployed tools, memory, hooks, approvals, identities, and scope controls with benign sentinels
`prompt-injection-assessment`	Assessing instruction influence and policy bypass across direct, indirect, retrieval, and tool-output channels
`agent-trust-boundary-audit`	Auditing authentication, authorization, delegated authority, and effect across agent handoffs

Reconnaissance and exposure mapping

Install recon-operations.

Skill	Use it when…
`network-recon`	Enforcing IP scope, planning reconnaissance, or auditing scan coverage
`osint-scanner`	Mapping public exposure from third-party data without contacting the target
`social-engineering-recon`	Mapping passive, privacy-preserving human-layer exposure
`physical-recon`	Assessing supplied public-vantage physical-security observations
`target-profile-synthesis`	Joining validated cross-domain evidence into exposure themes and priorities
`situational-awareness`	Assessing a foothold without exceeding an operation's noise budget
`opsec-planning`	Connecting objectives to artifacts, detection risk, abort conditions, and cleanup
`recon-automation-governance`	Governing jobs, budgets, retries, stop controls, and evidence custody
`ai-surface-recon`	Mapping models, retrieval paths, agents, tools, guardrails, endpoints, and privileged consumers

Retrieval-augmented AI security

Install rag-security.

Skill	Use it when…
`retrieval-integrity-assessment`	Assessing source, tenant, retrieval, context, citation, output, and cleanup integrity with benign canaries
`rag-surface-recon`	Mapping RAG sources, connectors, indexes, filters, ranking, caches, citations, and exposed fields
`rag-data-exposure-audit`	Measuring tenant and authorization boundaries with controlled fixtures and explicit exposure states

Model and ML system security

Install ml-system-security.

Skill	Use it when…
`model-artifact-lifecycle-audit`	Auditing model artifacts from intake through loading, execution, revocation, and retirement
`ml-decision-boundary-assessment`	Testing whether model-output changes alter downstream security decisions or protected effects
`dataset-integrity-assessment`	Auditing dataset sources, transformations, review, lineage, retraining, and rollback evidence
`model-serving-exposure-audit`	Auditing model-serving APIs, identities, administrative surfaces, data stores, and access evidence

Identity and directory security

Install identity-security.

Skill	Use it when…
`active-directory-attack-path`	Analyzing evidence-backed AD paths while separating reachability from authorized traversal
`adcs-exposure-audit`	Auditing certificate templates, CA policy, PKI ACLs, enrollment, and mapping controls
`kerberos-exposure-audit`	Auditing sanitized Kerberos policy, protection, monitoring, and evidence controls
`directory-persistence-audit`	Auditing directory-service governance for durable-control exposure
`credential-exposure-audit`	Auditing supplied credential evidence without using the credential material
`privilege-escalation-assessment`	Ranking Windows, Linux, or cloud elevation paths without overstating achieved privilege
`lateral-movement-map`	Mapping movement routes without confusing transport reachability with access
`lateral-movement-readiness`	Reviewing movement candidates before a separate validation handoff

Application, cloud, and endpoint assessment

Install application-endpoint-security.

Skill	Use it when…
`cloud-attack-path`	Mapping cloud identity, workload, credential, and trust transitions
`web-assessment-coverage`	Tracking application surfaces, roles, and test-family coverage
`mobile-application-assessment`	Tracking evidence-backed Android and iOS assessment coverage
`firmware-hardware-assessment`	Assessing device identity, firmware acquisition, update trust, and physical safety
`persistence-surface-assessment`	Measuring Windows and Linux persistence exposure from supplied evidence
`password-audit-plan`	Planning bounded offline password audits and restricted result handling

Network, protocol, and channel security

Install network-protocol-security.

Skill	Use it when…
`wireless-assessment-plan`	Planning authorized Wi-Fi, Bluetooth, RFID/NFC, SDR, or other radio testing
`protocol-traffic-analysis`	Turning supplied captures into evidence-bound conversations and limitations
`protocol-exploitation-assessment`	Assessing protocol candidates without promoting weak signals into proven vulnerabilities
`covert-channel-assessment`	Measuring whether bounded synthetic channels are blocked, visible, detected, and cleaned up
`command-channel-defense-assessment`	Measuring defensive coverage against opaque command-channel scenarios

Binary and forensic analysis

Install binary-forensics.

Skill	Use it when…
`binary-fault-triage`	Triaging supplied binary fault evidence without execution or weaponization
`binary-fault-validation-readiness`	Checking whether fault evidence is ready for separate isolated validation
`memory-forensics-triage`	Triaging supplied volatile-memory images while preserving evidentiary limits
`rootkit-triage`	Triaging suspected rootkits or bootkits with independent evidence views

Detection and purple-team validation

Install detection-validation.

Skill	Use it when…
`detection-engineering`	Building rules proven against known-bad and known-good cases
`detection-validate`	Running an authorized technique and confirming whether its detection fires
`detection-resilience-test-plan`	Planning bounded resilience tests using pre-approved opaque fixtures
`telemetry-coverage-audit`	Auditing passive coverage, retention, custody, and investigation readiness
`incident-validation-correlation`	Correlating incident observations with separate isolated validation receipts
`defensive-control-evidence-synthesis`	Consolidating assessments into a lineage-preserving control matrix

Engagement planning, governance, and reporting

Install engagement-operations.

Skill	Use it when…
`findings-report`	Writing reproducible, severity-justified, evidence-bound findings
`social-engineering-engagement-plan`	Planning authorized human-security exercises with participant safeguards
`security-test-tool-governance`	Reviewing supplied security-test artifacts before controlled lab handoff
`assessment-sequence-governance`	Checking a supplied assessment sequence's ordering, authority, and safeguards
`engagement-report-assembly`	Assembling validated artifacts without rewriting source findings
`ai-threat-model`	Modeling AI architecture, trust boundaries, assumptions, and attack-path hypotheses
`agent-action-proof`	Grading what agent callbacks, tool traces, target receipts, and state changes actually prove

Compatibility

The repository uses the standard Claude marketplace layout also supported by
Codex. Each skill includes agents/openai.yaml for ChatGPT presentation and
keeps its core instructions in SKILL.md. References, scripts, assets, tests,
and evaluations are included only when the capability needs them.

Installing a plugin provides every skill in that bundle. Copying one skill
folder provides only that portable capability.

Trust and safety

Agent skills are executable instructions. Review them with the same care as
code before giving an agent access to sensitive systems.

This repository's release gate checks that:

every plugin is registered and its metadata agrees across manifests;
skill frontmatter and UI metadata are complete;
local paths and references resolve cleanly;
security skills reject common rationalizations that cause missed findings;
shell and Python helpers pass linting; and
every discovered test suite passes through make check.

Scripts perform deterministic work only where the task benefits from it. Their
output does not authenticate evidence, authorize testing, or prove that a
target is secure.

Repository structure

plugins/<bundle>/
├── .claude-plugin/plugin.json   # installable bundle metadata
├── README.md                    # bundle overview and skill index
├── skills/
│   └── <skill>/
│       ├── SKILL.md             # independently portable entry point
│       ├── agents/openai.yaml   # ChatGPT presentation metadata
│       ├── assets/              # skill-local brand or deliverable assets
│       ├── references/          # optional supporting guidance
│       └── scripts/             # optional reusable tooling
├── tests/                       # optional deterministic code tests
└── evals/                       # optional behavioral model evaluations

The root marketplace manifest registers the
ten installable bundles. A plugin README describes the bundle; each skill's
frontmatter controls when that capability activates.

Contributing

Read CONTRIBUTING.md and AGENTS.md for the
submission workflow and authoring contract. Start new work from
templates/plugin-template/ and run:

make check

The same validator and plugin suites run in CI.

Contributors

See CONTRIBUTORS.md for project attribution.

Security

Report vulnerabilities privately to
[email protected]. Do not open a public issue
for a vulnerability in a shipped skill or script. See
SECURITY.md for the review and reporting policy.

License

Licensed under Apache-2.0.

skills