create-context-tests
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFEDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions involve running the
nao testandnao chatcommands within the local project environment to facilitate agent evaluation. - [INDIRECT_PROMPT_INJECTION]: The skill ingests external data in the form of user-provided SQL queries and natural language prompts to populate test templates.
- Ingestion points: Steps 1 and 3 involve receiving SQL and prompts from the user to create
tests/*.yamlfiles. - Boundary markers: The skill does not explicitly define boundary markers for the generated YAML content.
- Capability inventory: The
nao testutility executes the generated SQL queries against the data warehouse. - Sanitization: User-provided SQL is recorded into test files for execution without explicit sanitization steps.
- [CREDENTIALS_UNSAFE]: The skill references the use of an
ANTHROPIC_API_KEYand instructs the agent to allow the user to manually enter credentials when prompted by the CLI, which is a secure practice for managing secrets. - [DATA_EXFILTRATION]: The skill identifies the official Anthropic API endpoint (
https://api.anthropic.com/messages) for model interaction. This communication is consistent with the intended functionality of the agent tool.
Audit Metadata