The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the ingestion of untrusted user task descriptions and the analysis of existing application code during audits.\n
Ingestion points: The skill accepts plain-language task descriptions from the user to resolve development modes (in SKILL.md) and reads existing project files during accessibility and store audits (in references/audit.md).\n
Boundary markers: There are no specific delimiters or "ignore embedded instructions" warnings defined to separate user-provided content or external file data from the agent's core instructions.\n
Capability inventory: The skill utilizes powerful system tools including bash, run_command, write_to_file, replace_file_content, and grep_search (as defined in the agents/openai.yaml configuration).\n
Sanitization: The procedure does not include explicit validation, filtering, or escaping of the content ingested from user prompts or project files before it is processed by the agent.

mobile