PUA Breakthrough Mode
Security checks across static analysis, malware telemetry, and agentic risk
Overview
This is an instruction-only workflow that makes the agent more persistent and proactive, with no code, credentials, install steps, or hidden data access shown.
This skill appears safe to install as an instruction-only workflow. It is best for complex work where you want the agent to try multiple approaches before stopping. If you use it in a workspace with powerful tools, tell the agent what it may inspect or change, and set clear limits for risky actions.
Static analysis
No static analysis findings were reported for this release.
VirusTotal
VirusTotal findings are pending for this skill version.
Risk analysis
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 spend more effort exploring alternatives before concluding, which is intended but may be more assertive than the default behavior.
The skill deliberately changes the agent’s stopping behavior so it continues until explicit completion or blocker criteria are met. This is aligned with the stated purpose and includes stop boundaries.
Stop only when one of these is true: The completion criteria are met ... A blocking dependency ... The main paths have been tested ... Further exploration is lower value...
Use this skill for complex tasks where persistence is useful, and give clear stop conditions or limits when cost, time, or risk matters.
If the agent has access to tools, it may inspect files or take task steps sooner instead of only suggesting actions.
The skill encourages proactive use of available tools. This is purpose-aligned and bounded by a low-risk condition, but users should understand it may make the agent act more readily.
Default to action when tools are available and the risk is low.
For sensitive environments, state which tools, files, or actions are allowed before invoking the skill.
The agent may make reasonable next moves without repeatedly asking, which can be helpful but should be scoped for high-impact tasks.
The reference framework grants broad task-driving latitude, including tool use and reduced repeated permission checks. The surrounding framework still emphasizes evidence, constraints, and stop boundaries.
Let it inspect artifacts and run tools ... Let it propose fallbacks without waiting for repeated permission
Avoid using this mode for destructive, financial, legal, medical, deployment, or account-changing actions unless you provide explicit approval rules.