The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes test commands provided in the user input through spawnSync in bindings.mjs. It mitigates risk by using shell: false and enforcing a binary whitelist that includes python, python3, pytest, npm, pnpm, and node.- [INDIRECT_PROMPT_INJECTION]: The skill represents an attack surface for indirect prompt injection as it ingests untrusted data (goals and acceptance criteria) and processes it through a multi-stage agentic workflow.
Ingestion points: Goal and acceptance criteria provided at the start of the grill-test workflow.
Boundary markers: The workflow is strictly governed by application.json and application.mjs, which define specific transitions and validators for each stage.
Capability inventory: The skill has the capability to execute shell commands (spawnSync) and write files (fs.writeFile).
Sanitization: The skill implements a cleanEnvironment function to strip sensitive keys and tokens from the process environment before executing user-supplied commands.- [DYNAMIC_EXECUTION]: The skill utilizes a custom runtime environment (@rpamis/comet) to dynamically dispatch actions and sub-workflows based on the application manifest and current state.- [DATA_EXPOSURE]: The skill proactively implements security measures to prevent credential exposure by filtering the environment variables passed to sub-processes, specifically targeting keys related to tokens, secrets, and passwords.

grill-test