writing-offensive-skills
Skill401 starsChanged 56 days agoMITView on GitHub
Use when creating or editing a skill in this offensive-claude repo — for the SKILL.md conventions (trigger descriptions, technique map, runnable scripts, OPSEC/detection, red-flags tables, flowchart rules)
In this repository38 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- active-directory-attackSkill
- advanced-redteamSkill
- ai-agent-redteamSkill
- ai-securitySkill
- browser-exploitationSkill
- cicd-supply-chainSkill
- cloud-securitySkill
- coding-masterySkill
- container-k8s-escapeSkill
- crypto-analysisSkill
- edr-evasionSkill
- engagement-flowSkill
- engagement-memorySkill
- exploit-developmentSkill
- finding-disciplineSkill
- incident-responseSkill
- initial-accessSkill
- keylogger-archSkill
- malware-analysisSkill
- mobile-pentestSkill
- network-attackSkill
- opsec-disciplineSkill
- privesc-linuxSkill
- privesc-windowsSkill
- recon-osintSkill
- red-team-opsSkill
- reverse-engineeringSkill
- scope-disciplineSkill
- shellcode-devSkill
- threat-huntingSkill
- threat-model-disciplineSkill
- using-offensive-claudeSkill
- vulnerability-analysisSkill
- web-pentestSkill
- windows-boundariesSkill
- windows-mitigations-bypassSkill
- wireless-rfSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 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.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 53
Only for non-obvious decision points / "where you might stop too early".
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill 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
Writing Offensive Skills
Overview
Conventions for authoring skills in this repo so the dispatcher can find them and operators can trust them. This adapts superpowers' skill conventions to offensive security.
REQUIRED BACKGROUND: superpowers:writing-skills (the general conventions) and superpowers:test-driven-development (skills are tested like code — baseline failure first).
Description = triggers only
The description: field decides whether the dispatcher loads the skill. Write Use when…
triggering conditions and symptoms ONLY — never summarize the skill's workflow (a workflow summary
makes Claude follow the description instead of reading the skill).
# BAD (summarizes workflow): description: Recon skill that enumerates subdomains then scans ports
# GOOD (triggers only): description: Use when mapping a target's external attack surface — subdomains, hosts, exposed servicesThird person, technology-specific only if the skill is. Verb-first / gerund names.
Skill layout (progressive disclosure)
skills/<name>/
SKILL.md # thin router, <=180 lines
references/ # per-technique deep-dives (theory + 2024-2026 + code + detection + OPSEC)
scripts/ # runnable tooling (no placeholders)Domain (technique) skill SKILL.md sections, in order: frontmatter → When to Activate →
Technique Map (Technique | ATT&CK Txxxx | CWE | reference | script) → Quick Start →
OPSEC & Detection table → Deep Dives (links into references/).
Discipline skill (a hard rule, e.g. finding/scope/opsec-discipline): Overview with the Iron Law → the rule → Red Flags (STOP signals) → Rationalizations table (excuse | reality). State "violating the letter is violating the spirit" and close loopholes explicitly.
The four pillars (every technique)
2024-2026 currency (web-search-verified CVEs; no fabricated ids — mark unverified ones), runnable scripts, OPSEC + detection pairing, technique-level ATT&CK + CWE.
Flowcharts
Only for non-obvious decision points / "where you might stop too early". Never for reference material (use tables), code (use blocks), or linear steps (use lists).
Cross-references
Name only, with explicit markers: **REQUIRED:** scope-discipline. Never @-link (force-loads,
burns context). Frontmatter references:/scripts: list the files the skill ships.
Test before you trust
A skill that enforces discipline must resist rationalization under pressure. Capture the excuses an
agent makes without the skill, put each in the Rationalizations table, and re-check. Safety-relevant
scripts get a tests/ suite (run pytest) and an adversarial review before they're trusted.
Red Flags
- Description summarizes the workflow → rewrite to triggers only
- A domain skill with no Technique Map / no ATT&CK+CWE / no detection → incomplete
- A discipline skill with no Red Flags / no Rationalizations table → won't hold under pressure
- Fabricated CVE/arXiv ids presented as real → mark UNVERIFIED or remove
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.

