The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process construction project data from external files (CSV, Excel, JSON) for classification and analysis tasks, which constitutes a potential surface for indirect prompt injection.
Ingestion points: Project data and files processed through the methods defined in instructions.md and SKILL.md.
Boundary markers: While the skill does not define specific text delimiters, the instructions explicitly restrict the agent to 'Only use data provided by the user or referenced in the skill'.
Capability inventory: The skill is limited to file system read/write operations for data analysis; no dangerous primitives like arbitrary command execution or network exfiltration are present.
Sanitization: The instructions include a mandatory requirement for the agent to 'Validate inputs before processing', which serves as a primary security control.
[COMMAND_EXECUTION]: The skill requests filesystem permissions in claw.json to support its core functionality of reading project data and exporting analysis results to Excel, CSV, or JSON. The command usage is consistent with the skill's stated purpose, and no malicious command injection or unauthorized access patterns were detected.

few-shot-examples