The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it reads and processes markdown documentation files that could contain malicious instructions.
Ingestion points: The skill reads files from the content/ directory and referenced draft paths, as seen in the programmatic detection script using fs.readFileSync and the docs create command.
Boundary markers: No specific delimiters or warnings to ignore embedded instructions are provided in the instructions for the agent when processing these files.
Capability inventory: The skill allows for significant local command execution including yarn hugo, yarn test:links, node cypress/support/run-e2e-specs.js, and the docs CLI tool suite.
Sanitization: There is no evidence of sanitization or validation of the content read from markdown files before it is processed or used to inform subsequent actions.
[COMMAND_EXECUTION]: The skill facilitates the execution of various shell commands and scripts which, while standard for a documentation workflow, provide a mechanism that could be abused if the agent is misled by injected content.
Evidence: Instructions include running yarn, node, touch, and custom CLI tools like docs edit and docs create to manipulate the local filesystem and build environment.

content-editing