The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct the agent to install an external Python package (dos-kernel) from PyPI. While the author recommends pinning a specific version and provides warnings about typosquatting (e.g., distinguishing from the 'dos' package), the package originates from a community source outside of the pre-approved trusted organization list.
[COMMAND_EXECUTION]: The skill relies on the execution of shell commands (dos commit-audit, dos verify) to perform its primary function. This requires the agent to have shell access and the ability to execute third-party CLI tools.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data in the form of git commit subjects and bodies which could be authored by external contributors.
Ingestion points: Git commit history and diffs are read by the dos tool from the workspace.
Boundary markers: Not explicitly defined in the skill for isolating the tool's output from the agent's reasoning.
Capability inventory: The skill uses subprocess calls to execute the dos CLI and jq for parsing output.
Sanitization: The skill relies on the deterministic nature of the dos-kernel (stated to be non-LLM based) to prevent adversarial interpretation of commit data.

dos-verify-done-claims