The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions and documentation specify the use of uv run python to execute the local scripts scripts/pipeline.py and .codex/skills/unit-executor/scripts/run.py for workspace management and unit execution.
[DYNAMIC_EXECUTION]: The script scripts/run.py dynamically resolves the repository root by searching for a marker file (AGENTS.md) and prepends this path to sys.path to import the tooling.executor module from a computed location.
[INDIRECT_PROMPT_INJECTION]: The skill processes instructions and unit definitions from UNITS.csv and DECISIONS.md to determine which tasks to run and which sub-skills to invoke. This workflow lacks explicit sanitization for instructions embedded within these data sources.
Ingestion points: The skill reads from UNITS.csv, DECISIONS.md, and any files declared in the inputs field of a unit row.
Boundary markers: No specific delimiters or safety instructions are implemented to isolate untrusted unit metadata from the execution environment.
Capability inventory: The skill is capable of writing workspace files, updating status manifests, and executing other agent skills based on identity markers found in the manifest.
Sanitization: There is no evidence of validation or filtering for the semantic instructions or skill identities parsed from the project files.

unit-executor