The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted local data (CSV, JSON, etc.) for profiling and transformation, creating a surface for malicious content to potentially influence the agent.
Ingestion points: Data enters via the ae-cli data-integration inspect and convert commands which read local files specified by the user.
Boundary markers: The instructions mandate summarizing types and counts rather than printing raw values, though no technical delimiters are defined for the input data itself.
Capability inventory: The agent can execute arbitrary CLI tools, write to the local filesystem (handoff packages), and upload data to remote endpoints.
Sanitization: The agent is instructed to summarize samples, but there is no mention of sanitizing file contents before they are processed by the underlying tools.
[DYNAMIC_EXECUTION]: The "handoff" feature generates local scripts and configuration files that are executed in subsequent "reuse" sessions.
Evidence: The skill creates a package containing transform.mjs and various shell/python executors in a bin/ directory (e.g., run.sh, upload.sh).
Execution: Users are instructed to run these generated scripts to process similar-shaped files, which involves executing code generated by the agent's environment.
[COMMAND_EXECUTION]: The skill relies on executing various CLI commands (ae-cli, node, python, sh) to perform its primary functions.

ae-data-integration