The Agent Skills Directory

[COMMAND_EXECUTION]: Static analysis identified the use of subprocess.run within a shared utility script (_skill_cli.py). However, this function is not called by any of the scripts exposed as tools in the manifest, and the implementation correctly uses list-based arguments rather than a shell, mitigating command injection risks.
[EXTERNAL_DOWNLOADS]: The load_policy_contract.py script fetches configuration schemas from the vendor's official domain (https://cihook.heyeddi.com). This is a legitimate operation for retrieving live policy data and does not involve the execution of remote scripts.
[DYNAMIC_EXECUTION]: The skill uses runpy.run_path to handle deprecated script aliases and dynamically adjusts sys.path to import modules from a local vendor tool (heyeddi-ci) if present. These are controlled executions of local files used for tool integration.
[INDIRECT_PROMPT_INJECTION]: The inspect_repo tool reads file names and limited content from the project to identify development languages. While this exposes the agent to external data, the tool is read-only and lacks capabilities to execute data or write to the filesystem, neutralizing the impact of potential embedded instructions.

heyeddi-ci-runners