The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with the local system using standard CLI commands (omx --version, omx list, omx exec) to verify the environment and manage available tools. These actions are aligned with the skill's stated purpose of orchestration.
[EXTERNAL_DOWNLOADS]: Instructions for updates utilize standard package management and version control commands (npx, git) targeting the author's public GitHub repository. No suspicious or unverified remote sources were detected.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it processes external data through exploration and research tools.
Ingestion points: Codebase exploration via explore and iterative research via autoresearch (SKILL.md).
Boundary markers: The skill documentation references safety boundaries and re-orchestration triggers but does not provide specific implementation of delimiters in these files.
Capability inventory: The agent can execute commands via omx exec and manage subagents (references/capability-orchestration.md).
Sanitization: Guidelines explicitly prohibit the inclusion of credentials and session secrets in task handoffs (references/capability-orchestration.md).
[DATA_EXFILTRATION]: While the skill manages task orchestration and research, it explicitly implements rules to prevent the leakage of sensitive environment data and mandates human confirmation for high-cost or cross-domain integrations.

aha-codex-omx