The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md require the agent to execute a shell command involving user-provided variables (source-path and line). Without explicit instruction to escape or validate these inputs, there is a risk of command injection if the agent interpolates them directly into a shell environment.
[DATA_EXFILTRATION]: The script scripts/extract-example.mjs reads files from the filesystem using a path provided as an argument without validating that the path stays within the project workspace. This could be abused to read sensitive files if they contain JSDoc-formatted text.
[COMMAND_EXECUTION]: The script dynamically appends an Effect runner to the extracted code based on the user-supplied runner identifier. Although the identifier is validated with a regular expression, the practice of generating executable code from user-controlled parameters increases the risk of malicious code execution.
[PROMPT_INJECTION]: The skill processes untrusted content from external source files, creating a surface for indirect prompt injection. Ingestion points: File content at source-path read in scripts/extract-example.mjs. Boundary markers: None; the script extracts raw code blocks without adding security delimiters or warnings. Capability inventory: Filesystem read/write, shell execution, and reporting paths as clickable links. Sanitization: The script uses a slug function for filenames and a regex for identifiers, but does not sanitize the extracted JSDoc code content.

scratchpad