The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to execute several local commands and scripts, such as ./scripts/venv_setup.sh, python -m evals, nc, and docker compose. These are used for environment preparation, database connectivity checks, and running the test suite within the local development environment.
[EXTERNAL_DOWNLOADS]: The instructions reference external resources and documentation from docs.agno.com and os.agno.com. These are official domains owned by the skill author (agno-agi) and are used for providing technical guidance and platform integration.
[DATA_EXFILTRATION]: The workflow involves logging evaluation results to a Postgres database and provides instructions for connecting to the vendor's platform at os.agno.com to track performance history. This is documented as a standard feature for regression tracking.
[PROMPT_INJECTION]: The skill facilitates an Indirect Prompt Injection attack surface by processing external data as part of the evaluation suite.
Ingestion points: The agent reads and processes user-defined prompts and evaluation rubrics from evals/cases.py.
Boundary markers: Absent; the instructions do not prescribe specific delimiters or safety wrappers for the test case data.
Capability inventory: The agent has the ability to execute shell commands (Python/Docker) and modify source code files in the agents/ and evals/ directories.
Sanitization: No specific sanitization or filtering logic is mentioned for the processed case data.

eval-and-improve