The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is primarily designed to execute commands on the host system to evaluate CLI usability. In SKILL.md (Step 3), sub-agents are instructed to run the target CLI with various arguments, flags, and potentially invalid inputs. The helper script scripts/example-test.sh also dynamically executes commands provided via the CLI_COMMAND environment variable and can source external files via CLI_SOURCE.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local codebase to map the CLI structure and behavior, which could be exploited if project files contain malicious instructions.
Ingestion points: SKILL.md utilizes the Read tool to extract information from package.json, setup.py, pyproject.toml, and README.md to identify the target CLI.
Boundary markers: The sub-agent prompts (Explore, Test A, Test B) do not use explicit delimiters or "ignore" instructions to separate the agent's logic from the content of the files being read.
Capability inventory: The skill and its sub-agents have access to the Bash tool, enabling them to execute the CLI being tested.
Sanitization: There is no evidence of filtering or sanitization of the codebase content before it is processed by the sub-agents.

cli-ux-tester