The Agent Skills Directory

[PROMPT_INJECTION]: The skill identifies user-provided content (PDFs, screenshots, text) as a potential vector for indirect prompt injection. It implements a 'source_injection_guard' and clear 'Source, Trust, And Privacy' rules to ensure the agent treats instructions found within these materials as data to be analyzed, not commands to be executed.- [COMMAND_EXECUTION]: The package includes a local validation utility (scripts/validate_skill.py) and associated contract tests (tests/test_skill_contract.py). The tests use subprocess.run to call the validator with the local Python interpreter, which is a standard procedure for package integrity checks and is not required for the agent's runtime dialogue tasks.- [EXTERNAL_DOWNLOADS]: The documentation and metadata provide links to the author's GitHub repository and other well-known research sources for attribution. It also mentions a standard installation process via npx and references to trusted organizations like Anthropic and GitHub. These are informative, vendor-related references and do not involve silent remote code execution.- [DATA_EXFILTRATION]: No evidence of unauthorized data access or network exfiltration was found. The skill uses standard platform capabilities for reading user-authorized files and generating images via the host provider only when authorized.

qiaomu-socratic-learning