The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/load_policy_contract.py implements a discovery mechanism that searches for a local directory named heyeddi-ci in several relative locations (e.g., ../heyeddi-ci, ../../heyeddi-ci). If found, it modifies sys.path and dynamically imports the app.policy_insights module to execute code. This behavior allows for arbitrary code execution if a directory with the expected name and file structure is present in the search path.
[DYNAMIC_EXECUTION]: The deprecated script scripts/assert_runners_placeholder.py uses runpy.run_path to dynamically execute another script file (assert_runners_claims.py) at runtime.
[EXTERNAL_DOWNLOADS]: The scripts/load_policy_contract.py script fetches configuration data from https://cihook.heyeddi.com/api/public/eddi-ci-policy. While this targets a vendor-owned domain, it represents a runtime dependency on external content.
[INDIRECT_PROMPT_INJECTION]: The scripts/inspect_repo.py tool scans repository files to identify test commands and project metadata. This data is used to provide instructions to the agent, creating a surface where malicious content in files like package.json could influence agent behavior.
Ingestion points: scripts/inspect_repo.py reads package.json, pyproject.toml, README.md, and other project files.
Boundary markers: Absent in the tool's output to the agent.
Capability inventory: The agent is instructed to author and modify CI configuration files (eddi-ci.yaml, GitHub workflows).
Sanitization: Absent; content is extracted via regex and JSON parsing without escaping for prompt interpolation.
[COMMAND_EXECUTION]: The helper script scripts/_skill_cli.py contains a run_command function that utilizes subprocess.run to execute system commands. Although not directly invoked by the primary tools in the manifest, it provides a utility for shell command execution within the skill's environment.

heyeddi-ci-runners