The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the releases CLI and curl to interact with the vendor's API at api.releases.sh. These commands are integral to the skill's functionality for fetching release inputs and updating organizational overviews.
[CREDENTIALS_UNSAFE]: Authentication is handled via the RELEASES_API_KEY environment variable. The skill explicitly instructs the agent not to read sensitive information from .env files or other local sources, maintaining proper secret management practices.
[PROMPT_INJECTION]: The skill processes untrusted external release content, which is a potential surface for indirect prompt injection.
Ingestion points: Data is fetched through CLI tools or direct API calls as defined in the workflow (SKILL.md).
Boundary markers: The system prompt utilizes structured tags like <release-meta> and <existing-page> to isolate untrusted data (SKILL.md).
Capability inventory: Operations include shell command execution (releases, curl, jq) and file writes to the /tmp directory (SKILL.md).
Sanitization: Employs jq for secure JSON formatting and includes a specific system prompt directing the AI to treat input as data rather than instructions (SKILL.md).

regenerating-overviews