Security audit
Security checks for vulnerabilities and agentic risk
This is a Markdown-only skill-writing guide whose file-writing and install-location references match its stated purpose of helping users create Agent Skills.
Before installing, note that this skill is intended to help an agent create or edit skill files and explain how to place them in an agent skills directory. Use it when you actually want skill-building assistance, and review generated skills before installing or reloading them.
Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.
1. Gather Requirements → 2. Design Structure → 3. Write SKILL.md → 4. Add Supporting Files → 5. Validate → 6. Test
Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.
1. Gather Requirements → 2. Design Structure → 3. Write SKILL.md → 4. Add Supporting Files → 5. Validate → 6. Test
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.
1. Gather Requirements → 2. Design Structure → 3. Write SKILL.md → 4. Add Supporting Files → 5. Validate → 6. Test
Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.
After creating the skill:
1. **Place in correct location** for your agent (e.g. a personal or project-level skills directory such as `~/.claude/skills/skill-name/SKILL.md`, `.claude/skills/skill-name/SKILL.md`, or `~/.agents/skills/skill-name/SKILL.md` — follow your harness's convention).
2. **Reload the agent** so it discovers the skill.
Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.
After creating the skill:
1. **Place in correct location** for your agent (e.g. a personal or project-level skills directory such as `~/.claude/skills/skill-name/SKILL.md`, `.claude/skills/skill-name/SKILL.md`, or `~/.agents/skills/skill-name/SKILL.md` — follow your harness's convention).
2. **Reload the agent** so it discovers the skill.
The description says the skill should be used when "designing or refining skills, prompts, references, or supporting files," which is broad and lacks clear boundaries or exclusion conditions. This could match a wide range of everyday authoring requests, especially around prompts or supporting files, causing unintended invocation.
This markdown guide uses the example user input "Help me extract text from this PDF" as a trigger test, and earlier encourages including trigger keywords users would naturally say. The phrase "help me" is a very common conversational opener, so presenting it as part of trigger matching risks normalizing overly broad activators that could collide with ordinary requests.