The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/compile-combined.sh executes several system utilities including find, yq, python3, pandoc, and base64. It also uses a Python heredoc to dynamically execute parsing logic on manifest data.
[DATA_EXFILTRATION]: The orchestration script includes a --banner parameter that reads the content of a specified local file and encodes it into a Base64 data URL within the final HTML report. This functionality allows for the potential reading and exposure of sensitive system files.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection through its report aggregation process. \n
Ingestion points: The skill processes the suite.yaml manifest, suite-scope.md document, and output artifacts (HTML/Markdown) from multiple sibling skills. \n
Boundary markers: Only basic HTML character escaping is implemented for specific manifest fields. \n
Capability inventory: The compile-combined.sh script has permissions to read files, write to the local directory, and execute sub-processes. \n
Sanitization: Sanitization is limited to a simple character replacement function in the bash script, which may not be sufficient to prevent malicious content from sibling skills from influencing the final combined navigation artifact.

mapping-suite