The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill relies on external files to determine the sequence of actions to take, creating a vulnerability where malicious instructions in these files could influence agent behavior.
Ingestion points: Reads workflow definitions from docs/users/workflows.md, data/workflows.json, and references/workflow-cards.md.
Boundary markers: None identified. The skill does not provide instructions to ignore potentially conflicting or malicious prompts within these files.
Capability inventory: The skill has the ability to install other skills and orchestrate their execution, which can involve file system access, network operations, or shell commands depending on the sub-skills invoked.
Sanitization: None identified. The skill directly follows the steps outlined in the external documentation.
[COMMAND_EXECUTION]: The skill facilitates the installation of additional tools and components through a direct installer.
Evidence: Instructions include running a "direct installer" with specific IDs and destinations.
Mitigation: The skill follows safety best practices by requiring a --dry-run and explicit user authorization before proceeding with installations.

antigravity-workflows