Back to skill

Security audit

123skill-display-name

Security checks for vulnerabilities and agentic risk

Overview

The skill is a disclosed learning logger, but it also recommends broad persistent hooks, workspace memory promotion, and cross-session sharing that users should review before installing.

Install only if you want persistent agent learning/memory behavior. Prefer project-local .learnings files, review entries before promotion to AGENTS.md/CLAUDE.md/SOUL.md/TOOLS.md, avoid user-level global hooks unless you want every prompt covered, and never store secrets or raw transcripts in the learning logs.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (20)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The declared description is about recording lessons learned, corrections, failures, and improved approaches for future use. The supplied code does not capture, store, review, or manage learnings directly. Instead, it is a helper script that creates a new skill scaffold under a skills directory, writing a SKILL.md template and suggesting follow-up manual steps. While it references a 'learning entry' concept in comments and template text, its actual primary purpose is skill extraction/scaffolding, not continuous-improvement logging or review. This is a material purpose mismatch, and the code also performs undeclared filesystem creation behavior.

Content

No source excerpt is available for this finding.

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
91% confidence
Finding

Instructing users to place executable hook configuration in ~/.claude/settings.json creates persistent agent behavior from a high-trust config location. If misused, modified, or combined with unreviewed scripts, this can affect all future sessions and repositories, making compromise or data leakage durable and wide-reaching.

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 181)May include surrounding context.

sessions_send

Send message to another session:

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

Session Persistence

Medium
Category
Rogue Agent
Confidence
88% confidence
Finding

The skill encourages persistent storage under ~/.openclaw/workspace/.learnings, which can retain user-derived data across sessions. Session persistence is especially sensitive here because the same skill also discusses broad triggers and promotion into higher-level memory/context files, increasing the chance that incorrect, private, or adversarial content becomes durable state.

Content

Scanner excerpt · SKILL.md (reported line 81)May include surrounding context.

└── FEATURE_REQUESTS.md

text

### Create Learning Files

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

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill explicitly references cross-session tools such as session listing, transcript reading, and messaging between sessions. Even though it advises use only in trusted environments and with explicit user consent, exposing transcript access and inter-session sharing creates a real confidentiality boundary risk if the skill is activated too broadly or an agent misapplies it.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

Generic correction phrases like 'Actually...' or 'You're wrong...' are common in normal dialogue and can spur automatic logging when the user may only be giving transient feedback. Because the skill persists information to local files and may later promote it into broader memory files, accidental capture can create unnecessary data retention and prompt-injection persistence risks.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The feature-request trigger phrases are broad enough to match ordinary conversation, which can cause unintended activation and unsolicited logging. In this skill, unintended activation matters because activation can lead to filesystem writes and persistence of user content, creating privacy and integrity risks even without explicit malicious intent.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest claims this skill captures learnings, errors, and corrections to support continuous improvement. The 'Automatic Skill Extraction' section goes further by directing creation of entirely new skills via helper scripts and manual authoring workflows, which is a separate code-generation/capability-packaging function not justified by the stated purpose.

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
---

Session Persistence

Medium
Category
Rogue Agent
Confidence
86% confidence
Finding

Project-root .claude/settings.json establishes session-persistent hook behavior that automatically runs on future sessions in that repository. While useful operationally, persistence increases risk because broad hooks may continue injecting reminders or processing tool output long after the original need has passed.

Content

Scanner excerpt · references/hooks-setup.md (reported line 15)May include surrounding context.

Option 1: Project-Level Configuration

Create .claude/settings.json in your project root:

json
{

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

An empty matcher causes the UserPromptSubmit hook to run on every prompt, creating a broad, persistent trigger surface with no scope restriction. In a self-improvement skill, that increases the chance of capturing or influencing unrelated tasks across the entire session and can amplify prompt-injection or data-exposure risks.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

Using an empty matcher in a user-level configuration makes the hook activate globally across all prompts and projects. Because this persists in the agent's config directory, it broadens impact from one repository to every future session, increasing exposure of sensitive context and making unintended behavior harder to contain.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The 'minimal setup' still uses an empty matcher, so it remains always-on despite being presented as a lower-overhead option. That framing can mislead users into thinking risk is reduced when the trigger scope is still unrestricted.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The Codex CLI example also uses an empty matcher, enabling the hook on every prompt with no contextual limitation. Repeating this pattern across platforms normalizes overbroad agent instrumentation and increases the chance of unintended prompt capture or behavioral influence.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The document states the scripts 'only output text' and 'don't modify files or run commands,' but the hook configuration explicitly invokes shell scripts as commands. This is misleading security guidance because those scripts execute in the agent context with the user's permissions, and shell scripts can perform arbitrary actions if altered or replaced.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
88% confidence
Finding

The setup directs users to create persistent .learnings/ storage in the workspace or skill directory, enabling durable retention of agent-generated observations across sessions. In the context of a self-improvement skill, this increases the chance that sensitive prompts, operational mistakes, environment details, or user-specific information will be retained and later re-injected into future sessions, expanding both privacy and prompt-injection risk.

Content

Scanner excerpt · references/openclaw-integration.md (reported line 57)May include surrounding context.

openclaw hooks enable self-improvement

text

### 3. Create Learning Files

Create the `.learnings/` directory in your workspace:

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The document instructs the agent to persist "learnings" and promote them into long-lived workspace files, but it does not require sanitization, minimization, or exclusion of secrets, user data, or sensitive transcripts. In a self-improvement skill, failures and corrections often include command output, auth problems, file paths, prompts, or other sensitive context, so persistent logging can create an unintended data retention and prompt-injection surface.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill manifest describes a self-improvement capability focused on recording learnings, errors, corrections, and reviewing them before major tasks. This script instead scaffolds entirely new skill directories and SKILL.md files from a provided name, which is a repository authoring/generation function rather than a learning-capture implementation detail.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

Creating directories and templated SKILL.md files under ./skills gives this skill a code-generation and repository-modification capability. That capability is not obviously required to capture failures, user corrections, API issues, or outdated knowledge as described in the manifest.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
85% confidence
Finding

The skill directs creation of local directories and files during first use without an explicit warning that it will write to the workspace. While the writes are limited and not obviously malicious, silent persistence is still a real safety issue because it changes the user’s environment and may store sensitive conversation-derived content.

Content

No source excerpt is available for this finding.