The Agent Skills Directory

[COMMAND_EXECUTION]: The skill employs the Bash tool to run project-specific scripts, review-scope.py and qa-health-score.py, which are located in a shared scripts directory above the skill's base path. These scripts are used to define the boundaries of the quality assurance tasks and to generate health metrics.\n- [INDIRECT_PROMPT_INJECTION]: The skill interacts with and processes data from untrusted external sources, including web applications and their source code, which introduces a surface for indirect prompt injection.\n
Ingestion points: Untrusted content is ingested via browser interaction tools, WebSearch, and the Read tool while the agent performs testing and bug investigations as outlined in SKILL.md and exploration.md.\n
Boundary markers: The instructions do not include specific delimiters or directives to the agent to treat data from the application under test as non-instructional, increasing the risk that the agent might follow instructions embedded in HTML or code comments.\n
Capability inventory: The agent has access to sensitive capabilities, including the Bash tool for system commands and the Write and Edit tools for modifying the local filesystem.\n
Sanitization: No formal sanitization or filtering process is defined for the data retrieved from external web services or files before it is evaluated by the agent.

qa