Back to skill

Security audit

voice-setup

Security checks for vulnerabilities and agentic risk

Overview

This skill is mostly a writing-voice setup tool, but it needs review because it can persistently alter Claude Code behavior and can send draft passages with a dedicated model key to endpoints selected from repository files.

Review this before installing. Do not run install_host.py unless you want this skill to add persistent Claude Code output styles and hooks, and inspect the generated style plus claude_hook.py first. Avoid --apply-model in untrusted repositories; prefer an explicit endpoint you control, inspect any .rabbit-model file, and keep RABBIT_MODEL_API_KEY scoped to this tool only.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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 (2)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/rwlib/endpoint.py:474
Finding

Repository-Controlled Model Endpoint Can Exfiltrate Draft Passages and Dedicated API Credentials

Content
View full analysis
Remediation
View remediation

T01 · Skill Instruction Hijacking

Error
Location
scripts/install_host.py:313
Finding

Unvalidated Voice Markdown Is Persisted as Privileged Claude Code System-Prompt Content

Content
View full analysis
` in a profile, up to the next `## `.""" if not profile_md: return "" rx = re.compile(r"^##\s+" + re.escape(heading) + r"[^\n]*\n(.*?)(?=^##\s|\Z)", re.M | re.S) m = rx.search(profile_md) return m.group(1).strip() if m else "" ``` ```python one_line = section(profile_md, "Voice in one line") if one_line: lines.append("## In one line") lines.append("") lines.append(one_line) lines.append("") essentials = section(profile_md, "The three essentials") if essentials: lines.append("## First, before anything else") lines.append("") lines.append(essentials) lines.append("") ``` The resulting style is written persistently and selected in Claude Code settings: ```python voice, files = plan_install(args.scope) command = hook_command() if args.dry_run: prin ...[truncated 4056 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (75)

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.

Content

Scanner excerpt · SECURITY.md (reported line 74)May include surrounding context.

md
- It writes four paths and no others: two markdown files under
  `~/.claude/output-styles/`, the `hooks` and `outputStyle` keys inside
  `~/.claude/settings.json`, and a record of those at
  `~/.claude/rabbit-writes-host.json`.
- `--dry-run` prints every one of them and writes nothing, and the skill's
  own instructions require showing that output to the user before running the

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a voice-profile setup workflow, but the content documents materially broader behavior including scanning, rewriting, host hook installation, configuration changes, and remote model usage. This description-behavior mismatch can mislead users and orchestrators into granting trust or invoking the skill in contexts where they would not have consented to those capabilities.

Content

No source excerpt is available for this finding.