The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to access and source the ~/.zshrc file to retrieve the INTENSIVE_COLEARN_APIKEY. Sourcing a shell profile is a sensitive operation as it exposes the entire contents of the file, including all environment variables, aliases, and potentially other credentials, to the agent's execution context.
[COMMAND_EXECUTION]: The workflow requires the agent to execute shell commands, specifically source ~/.zshrc && export INTENSIVE_COLEARN_APIKEY, to prepare the environment for the bundled API client script.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes data from external API endpoints (intensivecolearn.ing). Content returned by the API (such as program descriptions or check-in records) could contain malicious instructions designed to influence the agent.
Ingestion points: External data enters the context via scripts/icl_api.py through various API operations like list-programs and get-program.
Boundary markers: The instructions include constraints to "Return only the fields needed for the task" and require user confirmation for mutating operations.
Capability inventory: The skill possesses network communication capabilities via urllib.request.urlopen and can read local files if specified through the --data-file argument.
Sanitization: The skill employs standard JSON parsing (json.loads) to process API responses.

intensivecolearn-api