usap-skills
CLAUDE.md4 starsChanged 2 months agoApache-2.0View on GitHub
81 standalone LLM skill packages + 13 cs-* orchestrator agents for the Unified Security Agent Platform (USAP). Each SKILL.md is a complete LLM system prompt. Skills are usable standalone (paste into any LLM) or as a USAP git submodule. 12 domains: appsec-devsecops, cloud-infra, detection, governance, identity-access, pentest, platform-ai, red-team, response, risk-compliance, system-security, webapp-security Key principle: Agents orchestrate skills; skills never depend on other skills. Two frontmatter formats coexist. The canonical format (used in templates/skill-template.md) is the new standard: Older skills…
In this repository100 other files it gives its agents.
- agents/CLAUDE.mdCLAUDE.md
- appsec-devsecops/CLAUDE.mdCLAUDE.md
- cloud-infra/CLAUDE.mdCLAUDE.md
- detection/CLAUDE.mdCLAUDE.md
- governance/CLAUDE.mdCLAUDE.md
- identity-access/CLAUDE.mdCLAUDE.md
- pentest/CLAUDE.mdCLAUDE.md
- platform-ai/CLAUDE.mdCLAUDE.md
- red-team/CLAUDE.mdCLAUDE.md
- response/CLAUDE.mdCLAUDE.md
- risk-compliance/CLAUDE.mdCLAUDE.md
- system-security/CLAUDE.mdCLAUDE.md
- webapp-security/CLAUDE.mdCLAUDE.md
- agent-integrity-monitorSkill
- ai-agent-securitySkill
- ai-ethics-governanceSkill
- ai-red-teamingSkill
- api-security-postureSkill
- appsec-code-reviewSkill
- appsec-customizeSkill
- attack-path-analysisSkill
- attack-surface-managementSkill
- behavioral-analyticsSkill
- build-integritySkill
- ciso-brief-generatorSkill
- cloud-security-postureSkill
- cloud-workload-protectionSkill
- compliance-mappingSkill
- container-image-scanSkill
- containment-advisorSkill
- continuous-pentestingSkill
- credential-attacksSkill
- cryptography-key-managementSkill
- cyber-insuranceSkill
- data-security-classificationSkill
- deception-honeypotSkill
- detection-engineeringSkill
- devsecops-pipelineSkill
- endpoint-os-securitySkill
- enterprise-risk-assessmentSkill
- findings-trackerSkill
- finding-triageSkill
- forensicsSkill
- guardrailSkill
- iac-securitySkill
- identity-access-riskSkill
- incident-classificationSkill
- incident-commanderSkill
- insider-physical-riskSkill
- internal-audit-assuranceSkill
- knowledge-managementSkill
- metrics-reportingSkill
- network-exposureSkill
- orchestratorSkill
- os-hardeningSkill
- ot-iot-device-securitySkill
- owasp-top10-classifierSkill
- patch-candidateSkill
- pentest-reportingSkill
- persona-coverage-auditSkill
- The first 60 shown.
The file
We split this CLAUDE.md into the rules it gives an agent
9 rules. About 0 are common, 1 is distinctive, and 8 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
8
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 117
Every skill tool and SKILL.md must produce a JSON payload with these required fields:
- Very specific: line 142
Commit messages: Conventional Commits — feat(skills):, fix(scripts):, docs(readme):, refactor(structure):, chore:
- Very specific: line 150
Check the next agent_id (run the grep command above)
- Very specific: line 152
Create the full package structure: SKILL.md, README.md, references/workflow.md, assets/templates/output-template.json, …
- Very specific: line 153
Ensure the SKILL.md body contains: Identity section, at least one classification/decision table with MITRE ATT&CK mappings, numbered …
- Very specific: line 155
Verify quality checklist in CONTRIBUTING.md before committing
- Distinctive: line 157
Do not place skills at the repo root: All skills live inside a domain subdirectory.
- Very specific: line 164
Add YAML frontmatter with cs- prefix, skills, domain, model: sonnet, tools
- Very specific: line 167
Update agents/CLAUDE.md agent catalog and root README.md agents table
≈ “Common” is an estimate: at least 1 in 5 of the 27,641 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
CLAUDE.md
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
What this repo is
81 standalone LLM skill packages + 13 cs-* orchestrator agents for the Unified Security Agent Platform (USAP). Each SKILL.md is a complete LLM system prompt. Skills are usable standalone (paste into any LLM) or as a USAP git submodule.
Repository layout
usap-skills/
├── <domain>/ # 12 domain directories (see below)
│ ├── CLAUDE.md # Domain-specific guidance
│ ├── README.md
│ └── <skill-slug>/ # Individual skill package
│ ├── SKILL.md # LLM system prompt + YAML frontmatter
│ ├── README.md
│ ├── references/workflow.md
│ ├── assets/templates/output-template.json
│ ├── expected_outputs/sample_output.json
│ └── scripts/<slug>_tool.py
├── agents/ # cs-* orchestrator agents
│ ├── CLAUDE.md # Agent dev guide — read before creating agents
│ ├── security/ # cs-security-analyst, cs-incident-responder, cs-red-teamer, cs-blue-team-analyst, cs-cloud-investigator, cs-supply-chain-defender, cs-threat-intel-lead, cs-purple-team-lead
│ ├── appsec/ # cs-appsec-engineer
│ ├── devsecops/ # cs-devsecops-engineer
│ └── executive/ # cs-ciso-advisor
├── templates/ # skill-template.md, agent-template.md, command-template.md
├── standards/ # frontmatter-spec.md, level-guide.md, naming-conventions.md, output-contract.md
├── shared/scripts/ # cvss_scorer.py, bb_scope_enforcer.py (no external deps)
├── domains/ # Domain index markdown files (one per domain)
└── references/ # Global reference docs12 domains: appsec-devsecops, cloud-infra, detection, governance, identity-access, pentest, platform-ai, red-team, response, risk-compliance, system-security, webapp-security
Common commands
# Run any skill tool (all tools support --output json)
python <domain>/<slug>/scripts/<slug>_tool.py --output json
# Run with input file
python <domain>/<slug>/scripts/<slug>_tool.py --input payload.json --output json
# Shared utilities (no deps required)
python shared/scripts/cvss_scorer.py --vector "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H"
python shared/scripts/bb_scope_enforcer.py --target example.com --scope-file scope.json
# Find the next available agent_id before creating a new skill
grep -r "^agent_id:" */*/SKILL.md | awk -F': ' '{print $2}' | sort -n | tail -1
# Validate frontmatter presence across all skills
grep -rL "^agent_slug:" */*/SKILL.mdArchitecture: Skills vs Agents
| Skill | cs-* Agent | |
| Location | <domain>/<slug>/ | agents/<domain>/ |
| Structure | Full package (SKILL.md + scripts + refs + assets) | Single .md file with YAML frontmatter |
| Purpose | Self-contained LLM prompt + CLI tool | Orchestrates multiple skills into a workflow |
| References other skills | Never (self-contained) | Via relative paths ../../<slug>/ |
| Naming | kebab-case slugs | cs- prefix required |
Key principle: Agents orchestrate skills; skills never depend on other skills.
SKILL.md frontmatter
Two frontmatter formats coexist. The canonical format (used in templates/skill-template.md) is the new standard:
---
name: <slug>
description: USAP agent skill for <Title>. Use for <one-line purpose>.
license: MIT
metadata:
version: 1.0.0
author: USAP Team
category: usap-detection # see standards/frontmatter-spec.md for allowed values
updated: YYYY-MM-DD
agent_slug: "<slug>" # quoted; must match name exactly
---Older skills use an extended frontmatter (with agent_id, level, plane, phase, ttl, approval_required, mutating_intents, can_execute, providers, required_invoke_role, required_approver_role, input_schema, output_schema, runtime_contract). Both are valid — do not convert old skills unless explicitly asked.
Skill levels
| Level | Role | Autonomy | Mutating? |
| L1 | Advisory / Board | Fully autonomous output | Never |
| L2 | CISO / Management | Autonomous analysis | Notifications only |
| L3 | SOC Analyst | Autonomous reads | Read-only tools |
| L4 | Technical / Expert | Supervised execution | Requires human gate |
Mutating intents (L4 skills that can recommend key rotation, isolation, etc.) must set human_approval_required: true in their output.
Output contract
Every skill tool and SKILL.md must produce a JSON payload with these required fields:
| Field | Type | Notes |
agent_slug | string | Must match directory slug |
intent_type | string | detect, respond, analyze, advise, escalate, report, or block |
action | string | Plain-English recommended next action |
rationale | string | Evidence-based explanation |
confidence | float | 0.0–1.0; below 0.5 = inconclusive |
severity | string | critical, high, medium, low, or informational |
key_findings | array[string] | At least one entry required |
evidence_references | array[object] | Required when severity >= high |
next_agents | array[string] | Downstream skill slugs (empty if terminal) |
human_approval_required | boolean | True for any mutating action |
timestamp_utc | string | ISO 8601 UTC |
Full schema and canonical example: standards/output-contract.md
Naming conventions (enforced)
- Skill slugs:
lowercase-hyphenated, max 4 words, no version suffixes - Script files:
<slug>_tool.py(hyphen in slug preserved, not converted to underscore) cs-*agents:cs-prefix, domain-scoped name- Commit messages: Conventional Commits —
feat(skills):,fix(scripts):,docs(readme):,refactor(structure):,chore:
Full spec: standards/naming-conventions.md
Adding a new skill
- Check the next
agent_id(run the grep command above) - Copy
templates/skill-template.md→<domain>/<slug>/SKILL.md - Create the full package structure:
SKILL.md,README.md,references/workflow.md,assets/templates/output-template.json,expected_outputs/sample_output.json,scripts/<slug>_tool.py - Ensure the SKILL.md body contains: Identity section, at least one classification/decision table with MITRE ATT&CK mappings, numbered Reasoning Procedure, explicit Intent Classification rule
- The Runtime Contract line must read:
../../agents/<slug>.yaml - Verify quality checklist in
CONTRIBUTING.mdbefore committing
Do not place skills at the repo root. All skills live inside a domain subdirectory.
Adding a new cs-* agent
- Copy
templates/agent-template.md→agents/<domain>/cs-<name>.md - Add YAML frontmatter with
cs-prefix,skills,domain,model: sonnet,tools - Reference skills via
../../<slug>/scripts/<slug>_tool.py(relative fromagents/<domain>/) - Minimum 3 workflows with concrete bash commands
- Update
agents/CLAUDE.mdagent catalog and rootREADME.mdagents table
Full guide: agents/CLAUDE.md
Shared utilities
Scripts in shared/scripts/ have no external dependencies. A script belongs here only if used by 3+ skills or provides core algorithmic capability (scoring, validation, classification). Single-skill helpers go in that skill's own scripts/ directory.
Domain CLAUDE.md files
Each domain directory contains a CLAUDE.md with: skills catalog, Python tools reference, SDLC/workflow integration, cascade intelligence (cross-skill routing), and domain-specific best practices. Read the relevant <domain>/CLAUDE.md before modifying skills in that domain.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

