The Agent Skills Directory

[COMMAND_EXECUTION]: The skill reads a YAML test plan from docs/testing/test-plan.yaml and executes shell commands specified in the setup.commands field. This allows for the execution of arbitrary commands defined within a local configuration file.
[COMMAND_EXECUTION]: During the test execution phase (Step 4b), the skill runs curl commands and agent-browser CLI actions using parameters (methods, URLs, headers, and bodies) sourced directly from the YAML test plan.
[DATA_EXFILTRATION]: The skill performs curl requests to URLs defined in the test plan. A maliciously crafted test plan could point these requests to external attacker-controlled servers to exfiltrate data.
[DATA_EXFILTRATION]: Upon test failure (Step 6), the skill executes git diff to identify changes and captures the output, along with potentially sensitive API responses, into a failure report file (docs/testing/evidence/<test.id>-failure.md).
[COMMAND_EXECUTION]: The skill has a significant attack surface for indirect prompt injection via the docs/testing/test-plan.yaml file.
Ingestion points: The YAML test plan is read in Step 1.
Boundary markers: None; the skill lacks delimiters or instructions to ignore embedded commands in the YAML.
Capability inventory: Execution of shell commands, background process management (nohup), network requests (curl), and browser automation (agent-browser).
Sanitization: While the skill uses yaml.safe_load to parse the file, it performs no validation or sanitization of the commands or URLs contained within it before execution.

run-test-plan