Back to skill

Security audit

ClawInboxRAG

Security checks for vulnerabilities and agentic risk

Overview

This skill is an email search helper, but it can run Gmail sync/indexing commands and relies on an unpinned external install path, so users should review it before installing.

Install only if you are comfortable granting the backend read-only access to Gmail and allowing local email indexing/embedding. Prefer a pinned release or reviewed commit, avoid dev extras for normal use, store tokens carefully, and treat mail sync as a state-changing maintenance action rather than a simple read-only search.

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
README.md:14
Finding

Unpinned External Repository and Dependency Installation

Content
View full analysis

Vulnerability Details

File Location: README.md, lines 14–18
Vulnerability Type: Unpinned third-party source and dependency installation
Risk Level: Medium

Vulnerable Code

bash
git clone https://github.com/dmoraine/ClawInboxRAG.git
cd ClawInboxRAG/skill
uv sync --extra dev

Technical Analysis

The documented installation process clones the repository's mutable default branch and immediately resolves and installs its project dependencies, including development extras. It does not pin a reviewed commit or signed release, verify a checksum or signature, or otherwise establish that the downloaded content is identical to the version reviewed during this audit.

The remote repository and its dependency metadata are outside the audited artifact. Consequently, their current or future installation behavior cannot be verified from this project. If the repository, maintainer account, release process, or dependency configuration is compromised, the instructions could cause users to install attacker-controlled code.

This is a supply-chain weakness rather than evidence that the currently audited scripts are malicious.

Attack Path

  1. An attacker compromises the referenced repository, its maintainer account, or a dependency declared by the remote project.
  2. The attacker changes the default branch or dependency metadata to include malicious installation or runtime code.
  3. A user follows the documented instructions and clones the mutable default branch.
  4. The user runs uv sync --extra dev, which resolves and installs the attacker-controlled project or dependencies.
  5. Malicious code executes during installation or when an installed component is subsequently invoked, subject to the behavior of the compromised package and tooling.

Impact Assessment

Successful exploitation could execute code with the privileges of the user performing installation. Depending on that user's access, the attack ...[truncated 433 chars]

Remediation
View remediation

Remediation Suggestions

  1. Pin installation to a reviewed immutable commit SHA or cryptographically signed release rather than the default branch.
  2. Publish expected checksums or signature-verification instructions and require verification before installation.
  3. Commit and enforce a dependency lockfile containing exact versions and integrity information.
  4. Avoid installing development extras in normal runtime setup; provide a minimal production dependency group.
  5. Use protected release branches, signed tags, mandatory review, and tightly controlled maintainer access.
  6. Add automated dependency and provenance scanning to the release process.
  7. Document a reproducible installation procedure that maps each Skill release to a specific reviewed backend revision.
Vulnerability Patterns
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (5)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The skill advertises a read-only, bounded local mailbox retrieval interface, but its documented behavior includes running backend maintenance/indexing commands like ingest-primary, embed, and refresh-labels. This mismatch is dangerous because operators may approve or invoke the skill under a low-risk mental model while it actually performs broader actions against a configured Gmail RAG repository, potentially altering local state, triggering large data processing, or accessing more email content than expected.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The allowlist includes ingest-primary, embed, and refresh-labels, which are operational or indexing actions rather than simple read-only retrieval. In the context of a skill advertised as bounded and read-only, exposing these commands can trigger state changes, background processing, or data propagation that exceeds the user's expected permissions and increases the blast radius of misuse.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
83% confidence
Finding

The skill declares executable behavior and dependence on environment variables, but does not define an explicit tool/permission scope. That omission weakens reviewability and containment because consumers cannot tell from the manifest what execution and environment access the skill expects, increasing the chance of over-broad runtime privileges being granted.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The wrapper behavior does not match the stated skill purpose: it executes gmail_rag.cli commands rather than a local mail ... retrieval backend. This mismatch is security-relevant because users and higher-level agents may trust the manifest description and invoke the skill under false assumptions, causing unintended access to a different codebase and data flow than advertised.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The script comments and description frame execution as safe, but the actual allowed subcommands include non-read-only operations. This kind of safety-claim mismatch is dangerous in agent ecosystems because it can lower scrutiny and cause automated systems or users to overtrust a wrapper that permits broader actions than represented.

Content

No source excerpt is available for this finding.