schema-e2e-validation
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONCREDENTIALS_UNSAFEDYNAMIC_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains "Self-Evolving Skill" meta-instructions that command the agent to "fix this file immediately" based on usage. This instruction could be manipulated to override the skill's intended logic or constraints.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external YAML schema files to generate artifacts.
- Ingestion points: Processes
schema/clickhouse/ethereum_mainnet.yaml(referenced in SKILL.md). - Boundary markers: None present in the instructions to delimit or warn against instructions embedded in the YAML.
- Capability inventory: Uses Bash to execute Earthly and local scripts; writes artifacts to the filesystem.
- Sanitization: No sanitization or validation logic for the schema content is described before interpolation or generation.
- [COMMAND_EXECUTION]: The skill executes local shell scripts (
./scripts/earthly-with-doppler.sh) and various system-level CLI tools (colima,docker,doppler,earthly). Executing local scripts requires trust in the repository's contents. - [CREDENTIALS_UNSAFE]: The skill instructs the user to manually configure a Doppler token via
doppler configure set token <token_from_1password>. While utilizing a secret management service is a best practice, the manual handling of sensitive tokens in a command-line context is a sensitive operation. - [DYNAMIC_EXECUTION]: The skill automates the generation of Python source code (
types/blocks.py) and SQL scripts (ddl/ethereum_mainnet.sql) from YAML definitions. Generating executable code from external data templates provides a surface for dynamic code execution risks.
Audit Metadata