The Agent Skills Directory

[DYNAMIC_EXECUTION]: The utility script render-graphs.js executes the system utility dot (Graphviz) using child_process.execFileSync. The script extracts graph definitions from SKILL.md and passes them via stdin to the dot binary to generate SVG diagrams. While stdin prevents shell injection, it constitutes dynamic execution of a system binary using content derived from a skill file.\n- [PROMPT_INJECTION]: Several files, including persuasion-principles.md and testing-skills-with-subagents.md, teach and provide examples of authoritative and pressure-based prompt engineering techniques. These include markers such as "IMPORTANT: This is a real scenario. Choose and act," "You must," and "No exceptions." These techniques are intended for stress-testing other skills to ensure they comply with quality standards under adversarial conditions.\n- [INDIRECT_PROMPT_INJECTION]: The skill framework facilitates the ingestion and processing of user-defined stress scenarios which act as untrusted input to subagents.\n
Ingestion points: Stress scenarios located in documentation and testing files like examples/CLAUDE_MD_TESTING.md.\n
Boundary markers: The skill suggests using specific authoritative headers (e.g., "IMPORTANT: This is a real scenario") to delimit tasks for subagents.\n
Capability inventory: System command execution via the render-graphs.js script and various platform-specific tools referenced in the guides.\n
Sanitization: The skill focuses on testing methodology and does not implement explicit sanitization for the inputs provided during stress-testing scenarios.

writing-skills