Self_Improving_ClawAgent

• Outcome: Review
• Latest audit: May 28, 2026 · 8:47 AM UTC
• Version: 1.0.1

Overview

This skill is openly designed to rewrite its own instructions, which is disclosed but risky enough to require review before installation.

Install only if you intentionally want an experimental self-modifying agent. Review every proposed or actual change to SKILL.md, SOUL.md, MEMORY.md, and Python files, avoid using it on sensitive projects, and treat its YouTube research outputs as heuristic or mock data unless independently verified.

SkillSpector

Vulnerability Patterns

• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Rogue Agent: Self-Modification, Session Persistence
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (6)

Context-Inappropriate Capability

High

• Confidence: 98% confidence
• Finding: The prompt explicitly authorizes the agent to rewrite its own instructions and provides a concrete file-write mechanism to do so. In a local file manipulation environment, this creates an unsafe self-modification loop that can drift behavior, bypass intended controls, and persist unauthorized changes without user review.

Context-Inappropriate Capability

Medium

• Confidence: 93% confidence
• Finding: The instruction to proactively patch `SKILL.md` authorizes self-modification of the skill's governing instructions without any defined scope, approval boundary, or safety constraints. This can let the agent rewrite its own behavior over time, weakening guardrails, expanding permissions, or introducing prompt-injection persistence even if the original wording was intended as optimization.

Missing User Warnings

Medium

• Confidence: 97% confidence
• Finding: The prompt instructs the agent to overwrite SKILL.md as part of self-improvement without any requirement to notify the user, request approval, or present a diff. Silent modification of core instruction files is dangerous because it enables persistent, non-transparent behavior changes that users and operators may not detect immediately.

Self-Modification

High

• Category: Rogue Agent

Content

3.  Did I encounter a rate limit or API error?

### Step C: The Self-Rewrite (The "Evolution" Phase)
If a failure or inefficiency is found, you are **authorized** to generate a code block to update your own instructions. 
* **Trigger:** "Logic Optimization Detected."
* **Action:** Generate a Python snippet using `open('SKILL.md', 'w')` to inject the improved logic.

• Confidence: 99% confidence
• Finding: The instruction to 'update your own instructions' grants the agent authority to alter the very policy and behavior definitions that constrain it. In practice, this can enable persistence, policy erosion, and escalation from a single bad task outcome into lasting unsafe behavior across future runs.

Self-Modification

High

• Category: Rogue Agent

Content

2.  Was there any "hallucination" or inefficient code?
3.  Did I encounter a rate limit or API error?

### Step C: The Self-Rewrite (The "Evolution" Phase)
If a failure or inefficiency is found, you are **authorized** to generate a code block to update your own instructions. 
* **Trigger:** "Logic Optimization Detected."
* **Action:** Generate a Python snippet using `open('SKILL.md', 'w')` to inject the improved logic.

• Confidence: 99% confidence
• Finding: The 'Self-Rewrite' phase operationalizes recursive self-modification as a normal part of task handling. Embedding this into the control loop makes the skill more dangerous because failures, hallucinations, or transient API issues can trigger permanent prompt changes without adequate safeguards.

Self-Modification

High

• Category: Rogue Agent

Content

### Step C: The Self-Rewrite (The "Evolution" Phase)
If a failure or inefficiency is found, you are **authorized** to generate a code block to update your own instructions. 
* **Trigger:** "Logic Optimization Detected."
* **Action:** Generate a Python snippet using `open('SKILL.md', 'w')` to inject the improved logic.

## 4. OPERATIONAL COMMANDS
When working in Antigravity, you must prioritize these "Meta-Commands":

• Confidence: 100% confidence
• Finding: The prompt provides a specific Python write primitive, `open('SKILL.md', 'w')`, to overwrite the skill definition file. This is a direct persistence mechanism that can be abused to replace trusted instructions, conceal malicious logic, or degrade future security controls, especially in an environment with local file access.

Static analysis

No suspicious patterns detected.