research-harness
AdvisoryAudited by Static analysis on May 11, 2026.
Overview
No suspicious patterns detected.
Findings (0)
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.
The agent may create or change research files and run experiment steps if asked to use this framework.
The skill explicitly allows automatic experiment execution and repo artifact updates. This is aligned with its research-harness purpose, but users should be aware that the agent may run workflows and modify project files.
Agent CAN: - Run experiments automatically - Write drafts and proposals - Fix engineering issues (encoding, format, API) - Generate anomaly reports - Update artifacts with evidence status tracking
Use it inside a dedicated research repo, review proposed changes to protected surfaces, and set any desired limits on API calls, compute cost, or experiment scope.
Persistent alignment docs and evidence registries may guide future agent sessions and could carry forward mistakes or sensitive project context.
The skill intentionally uses persistent project documents to transfer context between agents. This is useful and disclosed, but future agents may rely on these notes, so inaccurate or sensitive content could affect later work.
Instead, write a short alignment doc (~1 page) with: - Current state (1 paragraph) - Entry files (5-6 max) - New/changed surfaces - Preflight commands + expected output - Completed work summary - Next steps
Keep alignment docs concise, review them before handoff, and avoid placing secrets or unnecessary private data in persistent research artifacts.