The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to execute local shell commands for diagnosis, including cat .adlc/init-options.json to inspect configuration and command -v jq to check for dependencies.
[DYNAMIC_EXECUTION]: The diagnostic process involves running local bash scripts (.agents/skills/team-boot/scripts/boot.sh and scripts/acceptance-test.sh) and executing unit tests using pytest. These actions are confined to the project's local directory and are standard for troubleshooting the environment.
[INDIRECT_PROMPT_INJECTION]: The skill instructions involve the agent ingesting raw output from diagnostic commands. Evidence chain: (1) Ingestion points include command outputs from cat and bash scripts; (2) Boundary markers are largely absent, though check 3 mentions wrapping output in an EXTREMELY_IMPORTANT delimiter; (3) Capability inventory includes shell execution (bash, cat, pytest); (4) Sanitization is not performed as the skill explicitly requests raw, unscrubbed output for debugging. The surface exists but is consistent with the skill's diagnostic purpose.

team-diagnose