The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts to automate workflow management and summarize results.
Evidence: The skill instructs the agent to run scripts/new_workflow.py to create project structures, scripts/collect_results.py to aggregate work packet data, and scripts/verify_workflow.py for artifact auditing.
[SAFE]: Robust security controls and human-in-the-loop requirements are integrated into the skill's operational logic.
Evidence: references/risk-gates.md defines specific scenarios requiring user approval, such as file deletions, external system mutations, and access to sensitive data.
Evidence: The skill includes explicit instructions against storing credentials, secrets, or sensitive production data in reusable workflow artifacts.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection due to its data integration capabilities.
Ingestion points: scripts/collect_results.py reads and summarizes content from markdown files generated by sub-tasks or research packets.
Boundary markers: No explicit delimiters are specified for the integration of external packet results into the agent's summary.
Capability inventory: The skill has capabilities for file system modification and script execution.
Sanitization: The aggregation script filters for specific markers but does not perform active sanitization of the input text against instruction-based attacks.

codex-dynamic-workflows