The Agent Skills Directory

[DATA_EXFILTRATION]: Documentation examples demonstrate patterns for exposing sensitive local files and directory structures to remote cloud environments.\n
Evidence: references/images.md provides an example for mounting the ~/.aws credentials directory into a container image via .add_local_dir("/user/erikbern/.aws", ...).\n
Evidence: references/web-endpoints.md demonstrates an example of serving the root directory (/) of a remote container via an unauthenticated web server using python -m http.server -d /.\n- [COMMAND_EXECUTION]: The skill facilitates the execution of arbitrary Python code and shell commands on remote infrastructure as a core feature.\n
Evidence: Use of @app.function, modal run, and subprocess.Popen in examples throughout the skill documentation and reference files.\n- [PROMPT_INJECTION]: The skill provides patterns for building public web endpoints that ingest untrusted data, serving as a surface for indirect prompt injection.\n
Ingestion points: Web endpoints defined in references/web-endpoints.md (e.g., @modal.fastapi_endpoint() processing data: dict).\n
Boundary markers: Examples do not utilize delimiters or specific instructions to isolate data from processing commands.\n
Capability inventory: The skill documents extensive capabilities including remote code execution, persistent volume writes, and network operations.\n
Sanitization: Provided examples do not demonstrate validation, escaping, or filtering of data received from external API requests.

modal