The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process user-provided documentation, which serves as an attack surface for instructions embedded in external data.
Ingestion point: User-supplied documents passed to the agent for review or restructuring as seen in SKILL.md.
Boundary markers: The instructions lack explicit delimiters or warnings to ignore instructions within the processed documents, increasing the risk of the agent obeying instructions found inside user content.
Capability inventory: The skill can read files, write responses, execute a local bash script (scripts/classify-doc.sh), and invoke external skills (humanizer).
Sanitization: There is no evidence of sanitization or filtering applied to the input text before processing.
[COMMAND_EXECUTION]: The skill uses a local bash script (scripts/classify-doc.sh) to analyze document content. While the script performs benign keyword counting, the invocation of shell scripts on user-provided file paths carries a standard command injection risk if the filename is not properly handled by the agent context.
[EXTERNAL_DOWNLOADS]: The skill instructions prompt the agent to recommend and potentially invoke an external skill (blader/humanizer) which is not from a recognized trusted vendor. This introduces a third-party dependency risk, as the agent is instructed to suggest installation via npx skills add blader/humanizer if it is not already present.

diataxis-writer