The Agent Skills Directory

[PROMPT_INJECTION] (SAFE): The skill does not contain instructions to override agent behavior, bypass safety filters, or reveal system prompts. It serves to restrict agent actions rather than expand them.
[DATA_EXFILTRATION] (SAFE): While the skill mentions sensitive files (e.g., .env, .env.local), it explicitly forbids their modification and contains no patterns for reading or transmitting their contents over a network.
[COMMAND_EXECUTION] (SAFE): All included commands are standard, non-destructive git operations (e.g., git status, git log, git diff). It explicitly warns against and requires authorization for destructive commands like rm -rf or git reset --hard.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill processes untrusted external data in the form of git log outputs and filenames.
Ingestion points: Reads output from git log and git status (SKILL.md).
Boundary markers: None present for command output parsing.
Capability inventory: Includes git commit, git add, and git push (SKILL.md).
Sanitization: The skill proactively mitigates risk by instructing the agent to use double quotes for all file paths to prevent shell interpretation of special characters in filenames.

git-safety