fastly-fiddle
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes a helper script,
scripts/run-fiddle.sh, which automates the testing workflow by executing shell commands. The script uses standard tools likecurlandjqto interact with the Fastly Fiddle API. These operations are limited to the intended functionality of the skill. - [EXTERNAL_DOWNLOADS]: The skill performs network operations to
https://fiddle.fastly.dev. This communication is necessary for submitting VCL code for remote compilation and execution, and for receiving real-time test results from the Fastly edge infrastructure. - [INDIRECT_PROMPT_INJECTION]: The skill processes external responses from the Fiddle API, which introduces a surface for indirect prompt injection if the test output were to contain adversarial instructions.
- Ingestion points: Data is ingested from the
https://fiddle.fastly.dev/results/endpoint viacurlin thescripts/run-fiddle.shscript. - Boundary markers: The script does not wrap the API output in specific boundary markers or include explicit warnings to ignore instructions within the fetched data.
- Capability inventory: The skill possesses network access (via
curl) and the ability to execute shell scripts and process data usingjq. - Sanitization: The tool handles data primarily as structured JSON, but it does not sanitize descriptive text fields from the Fiddle execution results before presenting them to the agent.
Audit Metadata