multi-agent-performance-profiling
Warn
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill's methodology involves agents creating investigation scripts (such as
profile_pipeline.py) at runtime and executing them using theuv runcommand to perform instrumentation and phase-boundary timing. - [INDIRECT_PROMPT_INJECTION]: The workflow relies on an Integration Agent to read and synthesize findings from multiple other agents, creating a potential vector where data from one agent can influence the behavior of the synthesizing agent.
- Ingestion points: The integration agent reads reports (e.g.,
PROFILING_REPORT.md) and evidence files from subdirectories intmp/perf-optimization/created by other agents. - Boundary markers: Absent; there are no instructions to use delimiters or ignore instructions embedded in the reports during the synthesis process.
- Capability inventory: The agents have access to
Bashfor command execution andRead/Writetools for report and file management. - Sanitization: Absent; the content generated by investigation agents is processed directly without validation, escaping, or filtering before being synthesized into the master report.
Audit Metadata