Back to skill

Security audit

agentself

Security checks for vulnerabilities and agentic risk

Overview

The skill is coherent, but it handles wallets, secrets, and email while relying on unpinned CLI/tool installation and incomplete referenced workflows.

Review this before installing. Use it only if you trust the agentself package source, can verify the installed version, and are comfortable with persistent identity storage plus possible access to secrets, email workflows, and a live wallet. Avoid live wallet or secret operations until the CLI and fetched tools are pinned or otherwise verified.

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

Error
Location
SKILL.md:9
Finding

Unpinned Third-Party CLI Installation and Unverified Tool Retrieval

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 9 and lines 35–39
Vulnerability Type: Unpinned dependency installation and unspecified component retrieval
Risk Level: High

Vulnerable Code

bash
Requires the `agentself` CLI. If unavailable, install it with `uv tool install agentself` or `pipx install agentself`, then run the version check below.
bash
agentself install --tools
agentself init
agentself --json show

Technical Analysis

The Skill instructs the Agent to install the agentself package without specifying an exact version, cryptographic hash, lockfile, signed artifact, or verified package source. Consequently, the code executed during installation may differ from the code that was originally reviewed.

The installed CLI is subsequently authorized to run agentself install --tools, which may retrieve additional components. The Skill does not identify those components or specify their sources, versions, checksums, signatures, or required privileges.

Although the Skill checks the CLI schema and warns about unexpected executable paths, these checks occur after installation and do not establish the integrity of the package or its dependencies. A compromised package can execute arbitrary installation-time code before any version check takes place.

This is particularly sensitive because the declared CLI functionality includes access to identity data, secrets, email, and a live wallet capable of moving real funds. The available file does not directly instruct the Agent to transmit secret values, and there is no direct evidence of deliberate exfiltration. However, compromise of the installed dependency could enable such behavior.

Attack Path

  1. The Agent loads the Skill and determines that the agentself CLI is unavailable.
  2. Following the Skill instructions, it runs uv tool install agentself or pipx install agentself.
  3. The package manager resolves the latest ...[truncated 1497 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin agentself to a specifically reviewed version rather than installing an unconstrained release.
  2. Require cryptographic hashes for the package and all transitive dependencies through a reproducible lockfile or equivalent mechanism.
  3. Identify the authoritative package registry, publisher, and source repository, and verify package provenance before installation.
  4. Prefer signed, reproducible artifacts and verify signatures before executing installed code.
  5. Document every component retrieved by agentself install --tools, including its source, exact version, checksum, signature, purpose, and required permissions.
  6. Require explicit user confirmation before installing the CLI, retrieving additional tools, accessing email, or initiating live-wallet operations.
  7. Perform installation and initial validation in a sandbox with restricted filesystem, credential, process, and network access.
  8. Ensure integrity verification occurs before any package-controlled code executes; a post-installation schema check is not sufficient.
  9. Apply destination allowlisting and least-privilege controls to any network operations involving identity, email, wallet, or secret data.
  10. Include the referenced workflow documentation and audited CLI implementation in the review scope before authorizing sensitive operations.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (1)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The description uses broad activation language such as 'initialize or hand off an identity,' 'use its wallet or secrets,' and 'connect email, or process mail,' which can match many common user requests without strong scoping constraints. In this skill's context, that is more dangerous because it grants access to sensitive identity material, secrets, wallet operations, and email workflows, increasing the chance of accidental invocation and unintended exposure or high-risk actions.

Content

No source excerpt is available for this finding.