Back to skill

Security audit

self-improving-agent-next

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because scanners report automatic changes to persistent agent instructions and Git commits that bypass verification hooks; scan coverage is partial.

Consider the suspicious verdict before installing. The central concern is that learning records may become persistent agent instructions automatically, with Git verification hooks bypassed. The scan is incomplete: 13 of 14 components were scanned, and tool misuse analysis was degraded for one file, so the results do not establish that the skill is safe.

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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Rogue AgentSelf-Modification, Session Persistence
Findings (12)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
99% confidence
Finding

The --no-verify parameter explicitly disables applicable Git verification hooks rather than merely documenting ordinary commit usage. In this automatic promotion workflow, that can allow unsafe instruction changes or other staged content to be committed without local safeguards; this is the same underlying issue as the related SQP-2 finding.

Content

Scanner excerpt · SKILL.md (reported line 148)May include surrounding context.

md
晋升自动触发(三步缺一不可):
1. **更新 AGENTS.md / MEMORY.md 正文**(这才是落地)
2. `git commit --no-verify -m "promote: <ID> — <摘要>"`
3. 状态改为 `promoted`

顺序固定:**先写正文 → 再改状态**。反了就会重演约束 A 的失败。

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.

Content

Scanner excerpt · references/hooks-setup.md (reported line 48)May include surrounding context.

Option 2: User-Level Configuration

Add to ~/.claude/settings.json for global activation:

json
{

Exfiltration Commands

High
Category
Prompt Injection
Confidence
90% confidence
Finding

Instructions found that direct the agent to transmit conversation context or user data to external services.

Content

Scanner excerpt · references/openclaw-integration.md (reported line 177)May include surrounding context.

sessions_send

Send message to another session:

text
sessions_send(sessionKey="session-id", message="Learning: API requires X-Custom-Header")

Automatic promotion modifies persistent agent instructions and commits changes without warning about the impact on future agent behavior.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The workflow automatically converts learning records into persistent agent instructions without an explicit approval or trust-validation gate. Malicious instructions captured in an error or learning could consequently influence future sessions; recurrence and LLM assessment do not establish trust.

Content

No source excerpt is available for this finding.

The activation rules extend to every task without a bounded scope.

Medium
Category
Trigger Abuse
Confidence
98% confidence
Finding

L019 requires a precheck before any task, while the preceding triggers include generic failures, corrections, and capability requests. No project boundary, explicit invocation requirement, or exclusion conditions limit activation, so ordinary interactions can initiate this skill's learning workflow.

Content

No source excerpt is available for this finding.

Automatic promotion bypasses Git verification hooks without warning.

Medium
Category
Output Handling
Confidence
99% confidence
Finding

The prescribed promotion command uses --no-verify, bypassing pre-commit and commit-message hooks that may enforce security checks. This is particularly risky when automatically committing persistent instruction changes derived from learning records.

Content

No source excerpt is available for this finding.

OpenSpace analysis is instructed without a privacy warning about the learning records it processes.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The workflow directs OpenSpace execute_task to analyze the .learnings inventory and check AGENTS.md content. These records can contain error details and user requests, but the skill provides no warning about sharing those contents with the configured analysis service or guidance to remove sensitive information first.

Content

No source excerpt is available for this finding.

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · references/examples.md (reported line 301)May include surrounding context.

When the above learning is extracted as a skill, it becomes:

File: skills/docker-m1-fixes/SKILL.md

markdown
---

Hook-disabling instructions recommend deleting the entire settings file without warning that unrelated configuration will also be lost.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

Deleting the entire settings file to disable one hook also removes unrelated configuration and potentially other protective hooks or restrictions. The guide does not constrain this advice to a dedicated, disposable settings file or require a backup.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · references/openclaw-integration.md (reported line 62)May include surrounding context.

Create the .learnings/ directory in your workspace:

bash
mkdir -p ~/.openclaw/workspace/.learnings

Or in the skill directory:

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · references/openclaw-integration.md (reported line 68)May include surrounding context.

Or in the skill directory:

bash
mkdir -p ~/.openclaw/skills/self-improving-agent/.learnings

Injected Prompt Files

Cross-session handoffs require full context without a privacy warning.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The handoff rule requires full context without limiting it to information the receiving session needs or is authorized to receive. If the originating session contains credentials, private transcripts, or restricted project information, following this rule can propagate those details across session boundaries unnecessarily.

Content

No source excerpt is available for this finding.