The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes numerous shell commands to manage the development environment, git worktrees, and Android devices. Evidence includes the scripts/detect-env.sh script and instructions in SKILL.md that use git, adb, gradlew, npm, and the orca orchestration CLI. These operations are consistent with the intended purpose of the skill.
[INDIRECT_PROMPT_INJECTION]: The multi-agent orchestration workflow creates an attack surface where a sub-agent could potentially inject malicious instructions into its deliverables or completion reports. The Coordinator agent ingests this data via orca orchestration check and orca orchestration worker-read. While the skill mandates validation and uses a structured '7-Point Contract' (assets/worker-task-template.md), the 'Automatic Mode' feature—which merges tasks after a 2-minute user intervention window—reduces the opportunity for human review. The skill relies on the Coordinator's own logic for validation rather than programmatic sanitization or isolation of sub-agent outputs.
[DATA_EXPOSURE]: The skill manages environment configurations and caches architectural digests in ~/.config/android-skill-derek/. This is a standard practice for development tools. The instructions explicitly recommend using environment variables for signing keys instead of hardcoding them, which is a secure approach to secret management.

android-agent-orchestration