self-improving-robotics

• Outcome: Pass
• Latest audit: Aug 28, 2026 · 7:23 PM UTC
• Version: 1.1.1

Overview

This skill is a local robotics learning logger with optional reminders; the higher-risk hook behavior is disclosed and opt-in but should be enabled narrowly.

Install only in robotics workspaces where local learning logs are useful. Start without hooks or with the minimal project-level reminder hook, narrow the matcher to robotics-specific terms, avoid `PostToolUse` if command output may contain secrets, and review diffs before promoting learnings into agent docs or generated skills.

SkillSpector

Vulnerability Patterns

• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• MCP Least Privilege: Underdeclared Capability, Wildcard Permission, Missing Permission Declaration
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration

Findings (5)

Lp3

Medium

• Category: MCP Least Privilege
• Confidence: 83% confidence
• Finding: The skill documentation includes network-capable actions such as cloning from GitHub and installing via a package-like command, but there is no explicit permissions model or declaration warning operators that network access is involved. In agent environments, undeclared network behavior weakens trust boundaries and can lead to unreviewed remote content being fetched or installed.

Tp4

High

• Category: MCP Tool Poisoning
• Confidence: 93% confidence
• Finding: The stated purpose is passive logging of robotics learnings, but the skill also describes hook-based session injection, post-tool output scanning, and skill extraction that writes new content under a skills directory. This description-behavior mismatch is dangerous because users may grant trust to a logging skill that in practice modifies agent behavior, inspects command output, and scaffolds new artifacts with broader security implications.

Vague Triggers

Medium

• Confidence: 88% confidence
• Finding: The hook matcher uses broad terms like 'control', 'safety', and 'sensor', which can appear in many unrelated prompts and cause the skill to activate unexpectedly. Overbroad automatic activation increases prompt-surface exposure, creates noisy or unintended behavioral changes, and may trigger logging or reminders outside the intended robotics context.

Vague Triggers

Medium

• Confidence: 88% confidence
• Finding: The advanced hook example repeats the same broad matcher and adds PostToolUse automation, increasing the chance that unrelated Bash output will be inspected and processed. This expands the monitoring scope beyond the declared use case and can unintentionally capture sensitive operational data or alter agent behavior too frequently.

Skill Enumeration

Medium

• Category: Agent Snooping

Content

### Ownership Rules
- This skill writes only to `.learnings/robotics/` in stackable mode.
- It may read other skill folders for cross-linking, but should not rewrite their entries.
- Standalone mode writes to this project's `.learnings/*.md` log files only.
- Stackable mode writes only to the namespaced folder above and must not rewrite other skills' log entries.
- Promotion into `AGENTS.md`, `SOUL.md`, `TOOLS.md`, `MEMORY.md`, rules, hooks, or generated skills is not a logging write. Show a reviewed diff and apply only after explicit user approval.

• Confidence: 80% confidence
• Finding: The skill explicitly allows reading other skill folders for cross-linking, which can enable enumeration of installed skills, project structure, and adjacent operational context. In multi-skill agent environments, this broadens information access unnecessarily and may expose sensitive metadata or facilitate chaining with other skills' artifacts.

VirusTotal

64/64 vendors flagged this skill as clean.

Static analysis

No suspicious patterns detected.