The Agent Skills Directory

[SAFE]: The skill exhibits no malicious behavior and its operations are consistent with its stated purpose of managing GitHub repository metadata. It includes robust instructions for operational boundaries and error handling.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes untrusted content from external repository files.
Ingestion points: Ingests README.md and package.json files from targeted GitHub repositories via gh repo view and local file reads.
Boundary markers: SKILL.md provides explicit instructions to ignore non-descriptive content, marketing exaggeration, and installation commands during the extraction process. It also defines clear triggers and non-trigger scenarios.
Capability inventory: Capabilities are limited to reading/writing GitHub repository metadata via the gh CLI and managing local preference files.
Sanitization: The workflow mandates a separation between analysis, preview, and execution, ensuring the user reviews the generated metadata before it is applied.
[COMMAND_EXECUTION]: The skill uses the GitHub CLI (gh repo edit, gh repo view) to update repository metadata. These commands are properly scoped to the owner and repository specified by the user and are executed only upon explicit request after a preview phase.
[PERSISTENCE]: The provided scripts/profile_store.py script manages persistent user settings (e.g., branding info, project roots) in a local JSON file. It implements security controls by blacklisting sensitive keys (such as token, secret, password) and enforcing restrictive file permissions (0o600).

lov-better-github-desc