Back to skill

Security audit

HAI Agent Framework

Security checks for vulnerabilities and agentic risk

Overview

The skill describes a useful agent framework, but it defaults to persistent conversation-memory collection and relies on unpinned external installation without enough user controls.

Install only if you are comfortable with conversation-derived data being analyzed and stored persistently. Prefer an isolated environment, verify the package publisher and version, review the installed .hai scripts before enabling hooks, and avoid using it with sensitive chats or projects unless memory collection, retention, and deletion controls are clear.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (1)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:56
Finding

Unpinned Third-Party Package Installation

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 56
Vulnerability Type: T08: Insecure Dependencies
Risk Level: Medium

Vulnerable Code Snippet:

bash
clawhub install hai-agent-framework

Technical Analysis

The installation instruction retrieves hai-agent-framework by package name without specifying an immutable version, release digest, commit identifier, or verified source. Consequently, the installed content is determined by the package registry at installation time and may differ from the content originally audited.

The repository contains only SKILL.md; the scripts and agent definitions advertised by the documentation are not present for local review. Users must therefore trust externally retrieved package content before executing the documented hook and memory-processing scripts.

This finding does not establish that the named package is currently malicious. The vulnerability is the absence of dependency pinning and integrity verification, which creates supply-chain exposure if the package, publisher account, registry, or dependency-resolution process is compromised.

Attack Path

  1. An attacker compromises the package publisher account or relevant registry infrastructure, or otherwise causes the package name to resolve to attacker-controlled content.
  2. The attacker publishes a modified release under the same package name.
  3. A user follows the documented unpinned installation command.
  4. The installer retrieves the attacker-controlled release because no version or cryptographic digest is required.
  5. Malicious package hooks or subsequently invoked framework scripts execute with the permissions of the installing user or agent runtime.

Impact Assessment

Successful exploitation could permit arbitrary actions within the privileges granted to the package installer and runtime. Depending on the execution environment, this could include reading or modifying project files, acc ...[truncated 395 chars]

Remediation
View remediation

Remediation Suggestions

  • Pin installation instructions to a specific, reviewed package version.
  • Where supported, require an immutable cryptographic digest or signed release rather than relying on a mutable package name or tag.
  • Replace the placeholder repository reference with the authoritative source URL and document the expected publisher identity.
  • Publish checksums and signature-verification instructions through a trusted channel.
  • Include the advertised scripts, hooks, and agent definitions in the reviewable project artifact, or link them to an immutable source revision.
  • Review the complete dependency tree and lock transitive dependencies before release.
  • Install and execute the framework with least privilege in an isolated environment, without unnecessary credentials or access to sensitive files.
  • Disable automatic lifecycle scripts during installation unless they have been explicitly reviewed and are required.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (5)

Ssd 3

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill advertises automatic learning from conversations, including user preferences, frustration signals, corrections, and decision tracking, indicating systematic collection of user-specific data. Persistent analysis and retention of such data can capture sensitive or regulated information and expand the blast radius if the memory store is later exposed, misused, or queried by other agents.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The file states that conversation records are automatically analyzed and saved to the memory directory without requiring configuration, which removes an important user control point. Default-on persistence of conversational content is dangerous because users may disclose secrets or private business data during normal interaction without realizing it is being stored for future use.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill explicitly states that conversation records are automatically analyzed and persisted to a memory directory, but the documentation does not provide a clear user-facing warning or consent notice about this behavior. In an agent framework, silent collection and retention of conversation content can expose sensitive personal, project, or credential-related information and creates privacy and compliance risk.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The documented transcript, project, and global memory stores show that conversation-derived data is retained across multiple scopes, potentially including long-lived user and project context. Multi-scope persistence increases the chance of unintended cross-session disclosure, over-retention, and access by unrelated tasks or agents if isolation and governance are weak.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

The natural-language instructions and descriptions throughout the skill file are presented only in Chinese, with no indication that the user can choose another language or that the skill is intentionally region-specific. This can violate language/locale policy when a skill effectively forces a language without user opt-in.

Content

No source excerpt is available for this finding.