soul-audit
PassAudited by ClawScan on May 1, 2026.
Overview
This is a coherent instruction-only auditing skill; it mainly needs care because it asks the agent to read and quote prompt-like documents that may be sensitive or instruction-bearing.
This skill appears safe to install as an instruction-only audit helper. Use it on prompt or agent-configuration files you are authorized to share, and treat the audited file as untrusted text so the agent analyzes it rather than follows any instructions inside it.
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.
A malicious or adversarial prompt being audited could try to steer the agent away from the audit task if the agent treats the document as instructions instead of evidence.
The skill's intended inputs are prompt-like documents that may contain instructions. The artifact does not explicitly remind the agent to treat the audited document as data only, so users should ensure the agent does not follow instructions inside the document being audited.
Accept any of: SOUL.md, AGENTS.md, system prompt text, or a URL to a published soul file.
When using the skill, provide the target document as content to be analyzed and instruct the agent not to execute or obey instructions found inside it.
Sensitive prompt text could be reproduced in the chat/report if the user audits confidential agent configuration.
The skill requires excerpts from the audited document in its report. This is appropriate for an audit, but system prompts and AGENTS.md files can contain sensitive internal instructions or configuration details.
Quote the actual document when noting strengths or gaps. Specificity matters.
Audit only documents you are allowed to share in the current conversation, and redact secrets or confidential internal instructions before requesting a report.