The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external test logs and configuration files to generate Markdown and JSON reports. A compromised plugin or malicious test output could theoretically contain instructions designed to influence the agent's interpretation of results.
Ingestion points: The runner ingests user-provided config.json files, plugin artifacts (.tgz), and capture the standard output/error from test processes.
Boundary markers: The documentation explicitly instructs users to review snapshot differences manually and verify outcomes outside the agent rather than trusting keyword matches.
Capability inventory: The skill utilizes child_process.spawn to run Docker, Node.js, and package managers, and writes results to a temporary report directory.
Sanitization: The redactLogs function in scripts/docker-release-smoke.mjs provides defense-in-depth by masking Bearer tokens, API keys, and other common credential patterns in the final reports.
[COMMAND_EXECUTION]: The skill orchestrates complex testing workflows using npm, pnpm, and docker. These operations are well-defined and critical to the skill's purpose as a test runner.
The implementation uses spawn with shell: false and includes a validation layer (validateConfig) that checks versions and profile names against regular expressions, effectively mitigating standard command injection vectors.
[EXTERNAL_DOWNLOADS]: The skill downloads and installs toolchain components (pnpm, @deepseek-ai/dsh) and plugin artifacts during the smoke test process.
These downloads target well-known package registries and official vendor repositories, which is standard behavior for development and testing tools.

plugin-test