The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from an external predictions_json file. Specifically, it parses the response and question fields to identify classification errors. While the skill's primary function is analysis, these ingested strings are reflected in the output kpi_gaps.jsonl and could influence downstream agents in a pipeline if they are not properly sanitized.
Ingestion points: predictions_json file in SKILL.md and references/skill_info.yaml (read by gap_analysis tool).
Boundary markers: None specified in the instructions for the agent to ignore embedded instructions in the JSON content.
Capability inventory: File reading, directory creation (results_dir), and command execution within a Docker container.
Sanitization: The skill description mentions word-boundary matching for 'yes'/'no', which provides a primitive form of validation, but full sanitization of free-text fields is not explicitly documented.

tao-analyze-gaps-vlm-bcq