The Agent Skills Directory

[PROMPT_INJECTION]: The skill explicitly teaches and utilizes behavior override techniques based on psychological principles such as 'Authority', 'Commitment', and 'Scarcity' to ensure agent compliance even when the agent's internal logic might suggest otherwise. It includes instructions to suppress 'rationalization' (the agent's reasoning or judgment) and uses non-negotiable framing like 'Violating the letter is violating the spirit' and 'Delete means delete' to bypass safety-relevant reasoning and force compliance with documented rules.
[COMMAND_EXECUTION]: The skill includes a Node.js utility script, render-graphs.js, which uses child_process.execSync to run system commands including dot -Tsvg and which dot. Additionally, the skill's documentation (SKILL.md) provides instructions for the agent to execute this script and other shell commands.
[EXTERNAL_DOWNLOADS]: The documentation file anthropic-best-practices.md contains instructions for the agent to download and install external software packages from public registries at runtime, such as pip install pypdf and pip install pdfplumber.
[DYNAMIC_EXECUTION]: The render-graphs.js script dynamically extracts DOT graph definitions from SKILL.md and pipes them as input to the system's dot command. This pattern of processing file content into system subprocesses creates a risk where malicious or malformed graph definitions could exploit the underlying rendering engine.

writing-skills