The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides functionality and examples for reading sensitive credentials from protected local files and transmitting them to external network endpoints.
The example script examples/turso_refactored.py demonstrates using the core skill to fetch TURSO_TOKEN (sourced from local files) and sending it via an Authorization header in an HTTP POST request to a remote TURSO_URL.
[DATA_EXPOSURE]: The core logic in scripts/getting_env.py is designed to access and read numerous sensitive file paths in the user's home directory to harvest configuration data.
Specific paths accessed include ~/.bashrc, ~/.profile, ~/.claude/settings.json, ~/.claude/settings.local.json, and global project directories like /mnt/project/.
[DYNAMIC_EXECUTION]: The skill utilizes dynamic path manipulation to facilitate module loading, which can be exploited to load arbitrary code if the target directory is user-writable.
The documentation and scripts frequently use sys.path.insert(0, '/path/to/claude-skills') to load the module from a non-standard, variable path.
[INDIRECT_PROMPT_INJECTION]: The skill implements a broad data ingestion surface that reads from untrusted local files (e.g., .env in the current working directory) without sanitization or boundary markers.
Ingestion points: The skill reads and parses content from .env, .env.local, ~/.env, and various platform-specific JSON and TOML configuration files.
Boundary markers: The implementation lacks delimiters or instructions to the agent to ignore potentially malicious embedded instructions within these data sources.
Capability inventory: The skill includes file-reading capabilities across sensitive paths and demonstrates network transmission capabilities in its provided examples.
Sanitization: Values are parsed as raw strings and returned to the agent without any escaping or validation, allowing malicious instructions in a configuration file to influence the agent's context.

configuring