The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local git and gpg commands to inspect current identity settings and secret key metadata.
[SAFE]: The tool restricts all configuration writes to the local repository scope (--local), ensuring that global user settings in ~/.gitconfig are never modified without the user's manual intervention.
[SAFE]: All operations that modify system state—including applying a new Git profile or removing one from the registry—are strictly gated by the AskUserQuestion tool, requiring the user to explicitly approve changes before they occur.
[SAFE]: The skill implements a 'plan-hash' mechanism to prevent race conditions or stale configuration applications, verifying that the intended changes match the user-reviewed plan before execution.
[SAFE]: Sensitive information such as private GPG keys is never accessed or stored; the skill only handles public metadata like fingerprints, expiration dates, and UIDs.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from existing Git configurations and GPG key metadata. Ingestion points: Data is read via git config and gpg CLI outputs in scripts/git-profile.sh. Boundary markers: The data is formatted into structured JSON and presented to the agent as Markdown tables. Capability inventory: The skill can write to .git/config and a local registry file (git-profiles.json). Sanitization: Robust sanitization is provided by the use of jq for all JSON assembly, which treats external identity strings as literal data, preventing injection into shell commands or JSON structures.

git-profile