The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a specialized Python script and shell instructions to dispatch GitHub Action workflows and perform local comparison of evaluation metrics.\n- [EXTERNAL_DOWNLOADS]: Fetches structured evaluation results and metadata from the project's dedicated CDN at results.eval.all-hands.dev. These downloads are neutral and associated with the vendor's own infrastructure.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external evaluation reports and interpolates it into GitHub comments, which could lead to indirect injection if the source data is compromised.\n
Ingestion points: Evaluation reports and metadata are fetched in scripts/manage_evals.py from the results.eval.all-hands.dev CDN.\n
Boundary markers: Absent. The markdown report generated in scripts/manage_evals.py does not use delimiters to isolate ingested data.\n
Capability inventory: scripts/manage_evals.py can perform network operations via the GitHub API to trigger workflows and post PR comments.\n
Sanitization: Absent. The script extracts and processes data fields from the fetched JSON without specific validation or escaping measures.

manage-evals