sglang-bisect-ci-regression
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads source code archives directly from the vendor's official GitHub repository (sgl-project/sglang) using curl and extracts them for installation on remote targets.
- [COMMAND_EXECUTION]: The skill uses
sshanddocker execto run diagnostic and reproduction commands on remote servers based on user-supplied arguments (ssh_targetanddocker_container). - [DYNAMIC_EXECUTION]: The skill generates a custom Python reproduction script at runtime, transfers it to a remote target using
scp, and executes it within a Docker container to verify CI failures. - [INDIRECT_PROMPT_INJECTION]: The skill ingests and parses external data from GitHub Actions CI logs to extract failure signatures and drive the bisection process.
- Ingestion points: CI logs are retrieved through
gh run view --log(SKILL.md). - Boundary markers: None are defined to separate log content from agent instructions.
- Capability inventory: The skill has access to shell execution (
ssh,docker exec), network operations (curl), and source control (git,gh) (SKILL.md). - Sanitization: No filtering or sanitization of the log content is performed before processing.
Audit Metadata