The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies extensively on local shell commands executed via pnpm cli, pnpm build, and pnpm dev. These commands are used to exercise agent behavior, run integration tests, and manage local snapshots.
[PRIVILEGE_ESCALATION]: The instructions explicitly direct the use of sudo to install global NPM packages (e.g., npm i -g vercel) within a Docker sandbox to overcome permission errors (EACCES). It also involves verifying and trusting a custom CA certificate by setting the JUNIOR_DOCKER_SANDBOX_EXTRA_CA_CERTS environment variable to a path in /etc/pki/ca-trust/.
[EXTERNAL_DOWNLOADS]: The QA process involves downloading external tools via npm and performing network connectivity probes using curl against vendor-associated endpoints such as junior-dev.sentry.cool to verify OAuth callback flows.
[INDIRECT_PROMPT_INJECTION]: The skill describes an environment where the agent processes untrusted external data while maintaining access to powerful local execution capabilities.
Ingestion points: Targeted chat prompts (-p), interactive conversation inputs, mock reporting payloads from the dashboard API, and OAuth callback responses.
Boundary markers: The instructions do not specify any delimiters or warnings to ignore embedded instructions within these data sources.
Capability inventory: The agent has the ability to execute arbitrary shell commands via the pnpm cli wrapper, access local file systems, and perform network requests.
Sanitization: No explicit sanitization, filtering, or validation of the external payloads is mentioned before they are processed by the agent.

junior-qa