Huizai Context Optimizer
PassAudited by ClawScan on May 1, 2026.
Overview
This skill is coherent context-compression guidance with no evidence of malicious behavior, but it may preserve or evaluate sensitive conversation details as part of its intended purpose.
This appears safe to install for context-compression guidance. Before using it on confidential work, remember that summaries can preserve file names, errors, decisions, and next steps; do not include secrets, and be cautious with any optional workflow that sends compressed context to an external judge model.
Findings (2)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Compressed summaries may include file paths, implementation decisions, errors, and task history that could affect future agent behavior or expose sensitive project context if shared.
The skill intentionally preserves conversation state and artifact history in summaries. This is central to the stated purpose, but such summaries can retain sensitive details or become over-trusted in later work.
Maintain structured, persistent summaries with explicit sections for session intent, file modifications, decisions, and next steps.
Use the skill for intended context-compression tasks, but keep summaries task-scoped, avoid including secrets, and verify important file/action history before relying on it.
If the evaluation workflow is implemented with an external LLM judge, private conversation or project details in the compressed context could be transmitted outside the current agent session.
The evaluation framework describes sending compressed context and model responses to an LLM judge. This is disclosed and aligned with compression-quality evaluation, but it can involve sharing sensitive session context with another model/provider.
Feed probe question + model response + compressed context to judge
Only use approved judging providers, avoid sending secrets or confidential project content, and confirm where compressed context is processed before enabling external evaluation.