Back to skill

Security audit

Synomega Skill

Security checks for vulnerabilities and agentic risk

Overview

The skill appears to perform the chemistry tasks it claims, but its manual install path can fetch executable code from mutable remote sources without integrity checks.

Use the ClawHub install path rather than the README's raw curl commands. If installing manually, pin a reviewed commit and verify hashes for SKILL.md and synomega_run.py. Run the Python dependency in a dedicated environment, consider exact pinned requirements, and pre-fetch or mirror model assets for privacy-sensitive or reproducibility-critical use.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (2)

T03 · Remote Payload Retrieval and Execution

Error
Location
README.md:26
Finding

Unverified Remote Executable Retrieval from a Mutable Branch

Content
View full analysis

Vulnerability Details

File Location: README.md, lines 26–32
Vulnerability Type: Unverified retrieval of executable code from a mutable remote source
Risk Level: High

Vulnerable Code

bash
mkdir -p ~/.claude/skills/synomega
curl -fsSL https://raw.githubusercontent.com/zbc0315/synomega-skill/main/SKILL.md \
  -o ~/.claude/skills/synomega/SKILL.md
curl -fsSL https://raw.githubusercontent.com/zbc0315/synomega-skill/main/scripts/synomega_run.py \
  -o ~/.claude/skills/synomega/synomega_run.py

Technical Analysis

The manual installation procedure downloads Skill instructions and executable Python code directly from the mutable main branch of a personal GitHub repository. It does not pin a reviewed commit or release and does not verify a cryptographic signature or checksum.

Consequently, the code installed by a user can differ from the version that was audited. The downloaded synomega_run.py is intended to be invoked by the Agent, making the remote repository an effective post-review code-execution channel. The command also overwrites files at the destination without checking whether trusted versions already exist.

No malicious behavior was found in the bundled version of scripts/synomega_run.py. The risk arises from allowing its effective payload to change after review.

Attack Path

  1. An attacker compromises the upstream GitHub account, repository, maintainer credentials, or branch protection.
  2. The attacker modifies SKILL.md or scripts/synomega_run.py on the main branch.
  3. A user follows the documented manual installation procedure.
  4. curl downloads the modified content and overwrites the local Skill files without integrity validation.
  5. The Agent loads the altered instructions or executes the altered Python helper.
  6. The payload runs with the permissions and accessible resources of the user or Agent process.

Impact Assessment

A malicious replacement scr ...[truncated 560 chars]

Remediation
View remediation

Remediation Suggestions

  • Pin each download URL to a reviewed immutable commit hash or versioned release rather than main.
  • Publish SHA-256 hashes or signed release manifests and verify them before installing either file.
  • Prefer distribution through a signed, versioned Skill package instead of independent raw-file downloads.
  • Configure branch protection, mandatory review, signed commits or tags, and strong maintainer authentication.
  • Download into a temporary file, verify integrity, and only then atomically move it into the Skill directory.
  • Detect existing destination files and require explicit confirmation before overwriting them.
  • Document the exact reviewed version so users can reproduce the audited installation.

T08 · Insecure Dependencies

Warning
Location
README.md:37
Finding

Unpinned Executable Python Dependency and Transitive Supply Chain

Content
View full analysis

Vulnerability Details

File Location: README.md, line 37; SKILL.md, line 51
Vulnerability Type: Unpinned third-party executable dependencies
Risk Level: Medium

Vulnerable Code

README.md, line 37:

bash
pip install "synomega[gnn]"        # the package (neural backend)

SKILL.md, line 51:

bash
pip install "synomega[gnn]>=0.9.4"    # neural D-MPNN backend (torch) — recommended

Technical Analysis

The README allows any available version of synomega, while the Skill instructions allow any current or future release at or above version 0.9.4. Neither command pins an audited exact version, constrains transitive dependencies, or verifies package hashes.

Python packages and their build systems can execute code during installation, and imported packages execute code within the helper process. The unrestricted gnn extra also expands the transitive dependency set. A compromised upstream release, malicious dependency update, or unexpected future behavior can therefore enter the execution environment without changes to this repository.

No evidence was found that the currently referenced synomega package is malicious. This finding concerns the unsafe and non-reproducible dependency-resolution policy.

Attack Path

  1. An attacker compromises the synomega publishing account, one of its transitive dependencies, or the corresponding package build pipeline.
  2. A malicious or altered release is published to the configured Python package index.
  3. A user runs one of the documented unrestricted pip install commands.
  4. The resolver selects the compromised current or future release.
  5. Malicious code executes during installation, import, model loading, or a subsequent Skill operation.

Impact Assessment

A compromised dependency could execute with the privileges of the user performing installation or running the Skill. It could access files, environment variables, Python ...[truncated 436 chars]

Remediation
View remediation

Remediation Suggestions

  • Pin synomega to a specifically reviewed exact version, such as synomega[gnn]==X.Y.Z.
  • Pin all transitive dependencies in a lock file generated from a controlled environment.
  • Use a hash-locked requirements file and install with pip --require-hashes.
  • Install the dependency in a dedicated virtual environment with only the permissions required for local inference.
  • Review release notes and dependency changes before updating the lock file.
  • Use a trusted package index or controlled internal mirror and preserve reviewed package artifacts.
  • Align README.md and SKILL.md so both prescribe the same exact, audited dependency version.
  • Document and verify the integrity and provenance of automatically downloaded model and stock assets.
Vulnerability Patterns
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (7)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
78% confidence
Finding

The README instructs users to fetch executable skill content directly from raw.githubusercontent.com into the active skill directory using curl against the mutable main branch. That creates a supply-chain risk: if the repository, branch, or upstream account is compromised, users may silently install altered instructions or code that the agent will later trust and execute locally. The chemistry context makes this somewhat more sensitive because the skill can drive synthesis-planning workflows and may trigger local package/model downloads, but the core issue is generic remote-code trust rather than domain-specific abuse.

Content

Scanner excerpt · README.md (reported line 28)May include surrounding context.

bash
mkdir -p ~/.claude/skills/synomega
curl -fsSL https://raw.githubusercontent.com/zbc0315/synomega-skill/main/SKILL.md \
  -o ~/.claude/skills/synomega/SKILL.md
curl -fsSL https://raw.githubusercontent.com/zbc0315/synomega-skill/main/scripts/synomega_run.py \
  -o ~/.claude/skills/synomega/synomega_run.py

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
85% 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 · README.md (reported line 27)May include surrounding context.

Claude Code (manual)

bash
mkdir -p ~/.claude/skills/synomega
curl -fsSL https://raw.githubusercontent.com/zbc0315/synomega-skill/main/SKILL.md \
  -o ~/.claude/skills/synomega/SKILL.md
curl -fsSL https://raw.githubusercontent.com/zbc0315/synomega-skill/main/scripts/synomega_run.py \

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 · README.md (reported line 27)May include surrounding context.

Claude Code (manual)

bash
mkdir -p ~/.claude/skills/synomega
curl -fsSL https://raw.githubusercontent.com/zbc0315/synomega-skill/main/SKILL.md \
  -o ~/.claude/skills/synomega/SKILL.md
curl -fsSL https://raw.githubusercontent.com/zbc0315/synomega-skill/main/scripts/synomega_run.py \

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 · README.md (reported line 28)May include surrounding context.

bash
mkdir -p ~/.claude/skills/synomega
curl -fsSL https://raw.githubusercontent.com/zbc0315/synomega-skill/main/SKILL.md \
  -o ~/.claude/skills/synomega/SKILL.md
curl -fsSL https://raw.githubusercontent.com/zbc0315/synomega-skill/main/scripts/synomega_run.py \
  -o ~/.claude/skills/synomega/synomega_run.py

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 · README.md (reported line 29)May include surrounding context.

bash
mkdir -p ~/.claude/skills/synomega
curl -fsSL https://raw.githubusercontent.com/zbc0315/synomega-skill/main/SKILL.md \
  -o ~/.claude/skills/synomega/SKILL.md
curl -fsSL https://raw.githubusercontent.com/zbc0315/synomega-skill/main/scripts/synomega_run.py \
  -o ~/.claude/skills/synomega/synomega_run.py

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
88% confidence
Finding

The skill references and relies on multiple environment variables (SYNOMEGA_CACHE, SYNOMEGA_MIRROR, SYNOMEGA_MODEL, SYNOMEGA_STOCK, SYNOMEGA_FORWARD_MODEL, plausibility toggles) and describes automatic first-use network downloads, but it does not declare any explicit tool scope or permissions boundary. In a host that exposes environment or execution capabilities broadly by default, this can enable unintended access to local configuration, path injection into model/stock loading, or unreviewed remote fetching behavior beyond what a least-privilege skill should have.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
85% confidence
Finding

This markdown file documents that pretrained assets are downloaded automatically from GitHub or USTC on first use, which is a network operation that can affect privacy or system state. Although the behavior is described, there is no explicit warning or disclosure framing that the skill will make outbound network requests and write cache data locally.

Content

No source excerpt is available for this finding.