Back to skill

Security audit

Edit Skill

Security checks for vulnerabilities and agentic risk

Overview

This skill is a scoped helper for reviewing and editing agent skill files, and its sensitive behavior is disclosed and tied to that purpose.

Install this only if you want an agent to help maintain other agent skills. Prefer review-only mode for security-sensitive skills, and inspect diffs before accepting edits that touch safety rules, routing metadata, permissions, or tool instructions.

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
Vulnerability Patterns
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (6)

Self-Modification

High
Category
Rogue Agent
Confidence
95% confidence
Finding

This skill explicitly directs an agent to review and directly patch skill files, which is a self-modification capability. In an agent system, allowing a skill to rewrite skills can weaken safeguards, preserve or amplify bad instructions, or introduce persistent behavior changes that outlive the current request; the broad trigger language makes this more dangerous.

Content

Scanner excerpt · SKILL.md (reported line 6)May include surrounding context.

md
description: Use when asked to review, edit, tighten, shorten, deduplicate, clean up, or remove slop from an existing agent skill while preserving behavior. Use for skill files that need clearer triggering, shorter instructions, stale reference cleanup, host-compatible frontmatter, or fixes for vague, repetitive, outdated, unsupported, or hard-to-apply rules.
---

# Edit Skill

Use this for existing skill files. The job is to improve agent behavior, not to
make the prose sound nicer.

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

This skill is explicitly designed to review and modify existing agent skills, which is a form of self-modification or peer-skill modification. Even without overtly malicious instructions, allowing a skill to rewrite other skills can weaken safety boundaries, alter trigger behavior, or propagate insecure changes if invoked incorrectly or given adversarial input.

Content

Scanner excerpt · agents/openai.yaml (reported line 2)May include surrounding context.

yaml
interface:
  display_name: "Edit Skill"
  short_description: "Review and tighten existing agent skills while preserving behavior, trigger intent, safety boundaries, and host-compatible frontmatter."
  default_prompt: "When the user asks to edit, tighten, dedupe, clean up, or remove slop from an existing skill, classify review-only vs edit-in-repo, preserve the skill's real job and safety boundaries, cut vague or stale rules, and verify referenced paths or commands before reporting changes."
policy:

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · quick_validate.py (reported line 13)May include surrounding context.

python
OPENAI = ROOT / "agents" / "openai.yaml"

required_sections = [
    "# Edit Skill",
    "## Rules",
    "## Pass Order",
    "## Output",

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The invocation wording is broad enough to match many ordinary editing requests, which can cause the skill to activate outside its intended niche of editing agent skills. Over-broad triggering increases the chance of misrouting user tasks and applying self-modification guidance in inappropriate contexts, especially because this skill can patch files directly.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The default prompt uses broad, everyday phrases like 'edit,' 'clean up,' and 'remove slop,' which can cause the skill to be invoked in situations beyond narrowly scoped skill-maintenance tasks. That increases the chance of unintended activation on unrelated user requests, potentially applying repository-editing behavior or skill-rewriting logic in the wrong context.

Content

No source excerpt is available for this finding.