The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and update existing documentation and legacy CSV data, which represents an attack surface where embedded instructions could influence the agent's behavior.\n
Ingestion points: The skill reads README.md, scenario files in scenarios/, and bug reports in bugs/ during its bootstrap and update procedures. It also uses a script to read from a user-supplied CSV file.\n
Boundary markers: While it uses YAML frontmatter delimiters, it lacks specific prompts to ignore potentially adversarial instructions found within the body of the documents it processes.\n
Capability inventory: The skill possesses extensive file system capabilities (read, write, directory creation) and the ability to execute local scripts.\n
Sanitization: Content from files is used to drive logic (e.g., deduping, status updates) without explicit validation or sanitization of the text bodies.\n- [COMMAND_EXECUTION]: The skill requires the execution of included Python scripts (scripts/explode_state.py and scripts/materialize_state.py) via the command line to perform data transformations between CSV and Markdown formats. These scripts operate on paths provided by the agent or user.

qa-report