The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script plugins/env-helper/scripts/env_helper.py and recommends using subshell execution ($(python3 ...)) to resolve ecosystem constants. It also invokes the ls command to inspect local directory contents for configuration files.
[DATA_EXFILTRATION]: The skill is designed to access and retrieve highly sensitive data, specifically HuggingFace API tokens (HF_TOKEN) and environment configuration from .env files. Accessing these paths is considered a sensitive operation, although the skill includes explicit negative constraints intended to prevent the agent from outputting these values to the user chat.
[PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests untrusted data from environment files to determine system constants.
Ingestion points: The skill reads .env files at the root of the workspace using both the env_helper.py script and the ls -la .env command.
Boundary markers: The SKILL.md file includes a 'Token Leakage' negative instruction constraint to prevent the agent from repeating tokens in the chat window.
Capability inventory: The skill has the capability to execute subprocesses (python3, ls) and read/write local files as described in the fallback procedures.
Sanitization: There is no evidence provided that the content of the .env file is sanitized or validated before being processed by the agent or the helper script.

env-helper