The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Bash script (scripts/discover-docs.sh) to perform repository audits. The script is provided as part of the skill and employs standard system utilities like find and grep or ripgrep. It includes safety checks for Bash versioning and validates user-provided arguments such as file extensions and exclusion paths using regex and character filtering to prevent command injection.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze documentation files from a target repository, which represents a potential attack surface for indirect prompt injection.
Ingestion points: File content is read and searched via the discover-docs.sh script, which pipes line matches to the agent context.
Boundary markers: The instructions define strict formatting for findings and a specific 'Changeset' format for proposed mutations, which helps maintain context separation between instructions and data.
Capability inventory: The skill can search file contents and, with explicit human approval, perform documentation-scoped file operations such as creation, updates, and deletion.
Sanitization: While no automated natural language filtering is applied to the analyzed content, the skill enforces a mandatory human authorization gate before any mutations are applied to disk, serving as the primary control against unauthorized actions directed by untrusted data.

ssot