Back to skill

Security audit

Semantic Grep

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent local semantic code-search helper, but its documentation exposes a credential-like ClawHub token and installs unpinned remote packages.

Review carefully before installing. Do not reuse the embedded ClawHub token; it should be revoked and replaced with user-provided authentication. Prefer installing a pinned, reviewed semgrepll version with hashes, and be aware that indexing code creates local cached embeddings of project content.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (2)

T09 · Insecure Skill Coding Practices

Warning
Location
README.md:43
Finding

Hardcoded ClawHub Access Token in Installation Documentation

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
clawhub.yaml:18
Finding

Unpinned and Integrity-Unverified PyPI Dependency Installation

Content
View full analysis
Remediation
View remediation
``` ```bash python -m pip install --require-hashes -r requirements.txt ``` 3. Lock and review transitive dependencies rather than protecting only the top-level package. 4. Verify release provenance, repository ownership, signatures or attestations, and correspondence between the reviewed source revision and published package. 5. Prefer installation from a controlled internal package mirror after security review. 6. Update all installer files consistently so no supported installation path bypasses the pinned version and hash requirements. 7. Document whether model acquisition or optional backends make network requests, and avoid claiming fully offline behavior unless this has been verified for the pinned release and all supported configurations. ]]>
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (6)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The README publishes a live-looking installation token directly in a command, which may enable unauthorized registry access, token abuse, or impersonated installs if the token is valid. In a skill distribution context, embedded credentials are especially dangerous because they are easy to scrape, reuse, and propagate into automated agent workflows.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The README makes a strong '100% offline capable' claim while also instructing users to install via a remote registry using a tokenized command. This is misleading security messaging: users may underestimate network exposure, trust boundaries, and supply-chain risk during installation, especially in an agent-skill context where remote installs can pull unreviewed content.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The trigger phrases are broad enough to match ordinary user requests about searching code or indexing a project, which can cause the skill to activate unexpectedly. In an agent setting, this increases the chance of unintended command execution against local project paths or repositories without clear user intent confirmation.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The trigger keyword 'semantic search' is broad enough to match many ordinary user requests that are not clearly intended to invoke this specific skill. Overbroad activation can cause unintended skill routing, which may run commands or influence responses in contexts where the user did not explicitly request this package.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The phrase 'search code by meaning' is ambiguous and conversational, so it may overlap with general programming help requests rather than explicit invocation of this skill. In an agent environment, ambiguous triggers increase the chance of accidental activation and unintended execution of indexing or search-related commands.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The trigger 'find code semantically' is similarly broad and could match common natural-language requests about understanding or locating code. Because this skill exposes command execution paths like indexing and searching local projects, loose activation boundaries raise the risk of unintended tool use on the wrong project or without clear user intent.

Content

No source excerpt is available for this finding.