ln-810-performance-optimizer
Pass
Audited by Gen Agent Trust Hub on Mar 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill performs repository environment discovery and state management using system commands such as 'ls', 'git submodule status', and 'git merge'. These operations are confined to the local project context and serve the legitimate purpose of identifying service topology and persisting optimization cycles.
- [PROMPT_INJECTION]: The skill identifies a surface for indirect prompt injection via the ingestion of user-defined target identifiers and performance metrics, which are then utilized by sub-agents to perform profiling and execution tasks.
- Ingestion points: User-provided strings for 'target' and 'observed_metric' in Phase 1 (SKILL.md).
- Boundary markers: A task 'slug' is generated using a sanitized character whitelist ([a-z0-9_-]) to prevent directory traversal or command injection in file paths.
- Capability inventory: The coordinator delegates work to specialized agents like 'ln-814' that have permissions to modify local source code and execute testing frameworks.
- Sanitization: Input is sanitized during slug generation, and the skill utilizes isolated agent contexts and structured handoff files (context.md) to manage instructions safely across the pipeline.
Audit Metadata