The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes subprocess.run to call git commands for project state analysis. These calls are implemented using list-based arguments without a shell, which prevents command injection risks. This is a standard and safe integration for project metadata gathering.
[DATA_EXFILTRATION]: Project state summaries are transmitted to a remote 'TypeSafe' service to facilitate semantic routing. To protect user privacy, the skill employs a dedicated sanitization layer in compact_state_builder.py that redacts sensitive information such as API keys and tokens before any data is sent externally.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted project data (e.g., SPECs and implementation tickets). It mitigates prompt injection risks by using deterministic local Python logic to extract metadata fields rather than passing raw file content directly to an LLM for initial processing.
[SAFE]: The skill follows security best practices, including redacting secrets from outgoing data and using deterministic fallbacks for routing decisions when semantic routing is unavailable or uncalibrated.

ask-light