The Agent Skills Directory

PROMPT_INJECTION (HIGH): High vulnerability to indirect prompt injection (Category 8) due to unrestricted codebase analysis combined with file-modification capabilities. Ingestion points: The skill reads arbitrary project files via scout agents and 'actual source files' during the documentation update phase. Boundary markers: None identified; the skill does not use delimiters or instructions to ignore embedded commands in the source code it reads. Capability inventory: Extensive file-write access to README.md and multiple files within the docs/ directory. Sanitization: None present; the agent is instructed to directly verify and incorporate code evidence into its documentation output.
COMMAND_EXECUTION (LOW): The skill executes ls -la to map the directory structure. This is a standard reconnaissance command used to identify project targets for subsequent reading and writing operations.

docs-update