ClawHub
Back to skill

Security audit

Skill Polisher

Security checks for vulnerabilities and agentic risk

Overview

This is a markdown-only skill-polishing helper whose file changes are disclosed and gated on user approval.

Installers should understand that this skill can propose edits across a single skill or multi-skill bundle. Review the generated audit report and approve overwrites only when the target is really a ClawHub skill source you intended to polish.

A.I.G

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (6)

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · references/audit-guide.md (reported line 65)May include surrounding context.

md
**How to check:** Compare the original against the polished version. For each category above, identify whether the original contained instructions in that category. If it did, verify the polished version preserves the substance of those instructions — either in the SKILL.md or moved to a references/ file.

**Skip this section entirely if the original SKILL.md contains no behavioral or output instructions.** Not every skill has them. This check only applies when the original has content that falls into these categories.

### Content Accounting

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/rules.md (reported line 47)May include surrounding context.

Bad:

text
The setup process requires you to first obtain a GitHub Personal Access Token
by navigating to github.com/settings/tokens, creating a new fine-grained token
with public repository read-only access, and then passing it to the setup command.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The trigger examples are broad natural-language phrases such as 'Polish my SKILL.md' and 'Clean up these skills', which can overlap with routine editing requests and cause the skill to activate when the user did not specifically intend to invoke this specialized workflow. Unintended invocation can lead to inappropriate file rewrites, reference-file restructuring, or misleading transformations of skill content, especially because the skill is designed to modify publication-facing instructions.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
85% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · references/audit-guide.md (reported line 39)May include surrounding context.

md
- [ ] Post-flow output still present (end-of-turn sections, email draft offers, optional next-step prompts)
- [ ] Root routing logic still present (bundle mode — routing table, dispatch rules, infrastructure overview)

**Why this matters:** The LLM may skip reference files to save tokens. If these rules are in references/, the agent will hallucinate data, skip confirmations, or produce broken output. See rules.md for the full protected content table.

### LLM Execution

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

This markdown file states that the polisher 'auto-detects bundle mode — no configuration needed' without defining a narrow invocation boundary or explicit negative conditions beyond later heuristics. That broad activation rule could cause unintended bundle-mode behavior when a directory incidentally matches the listed patterns.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This markdown file explicitly instructs authors to include trigger phrases and then provides examples like 'How's my project doing?' and 'Give me a growth digest,' which are broad conversational phrases rather than narrowly scoped activators. Because the file is a rules/reference document for skill authoring, this guidance can propagate ambiguous triggers into downstream SKILL.md files.

Content

No source excerpt is available for this finding.