The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection via the coverage report file.
Ingestion points: scripts/check-coverage.sh reads data from coverage/coverage-summary.json in the local workspace.
Boundary markers: Absent. Values from the JSON file are interpolated directly into output strings without delimiters.
Capability inventory: The skill registers a PostToolUse hook on Bash commands, allowing it to provide feedback to the agent after testing commands like jest or vitest.
Sanitization: None. The script does not validate or escape the contents of the coverage report before echoing it to the console, which allows a malicious report to influence agent behavior.
[COMMAND_EXECUTION]: The script uses node -e to execute a JavaScript snippet for parsing the JSON coverage file. While the file path is hardcoded, the use of a runtime to execute code for data parsing represents a larger attack surface than static alternatives like specialized JSON parsers.

coverage-guard