The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes npx to download the @web42/stask package from the npm registry. This package originates from an unknown organization, @web42, which is not recognized as a trusted vendor.
[REMOTE_CODE_EXECUTION]: Execution of the @web42/stask package via npx constitutes remote code execution of unverified third-party code at runtime.
[COMMAND_EXECUTION]: The skill instructions require the agent to execute local shell commands, such as starting development servers and running test suites, based on specifications found in external task files.
[DATA_EXFILTRATION]: The skill is designed to capture and transmit local environment data, including screenshots and logs, to an external service through the qa command functionality.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8).
Ingestion points: Acceptance criteria and task details are fetched from external sources using the heartbeat and show commands (SKILL.md).
Boundary markers: Absent. The skill provides no instructions to isolate or disregard commands embedded within the external specification data.
Capability inventory: The agent has permission to execute shell commands, access the local filesystem, and send data to external channels (SKILL.md).
Sanitization: None. The agent is directed to follow the criteria in the external spec systematically, potentially leading to the execution of instructions disguised as test steps.

stask-qa