ClawHub
Back to skill

Security audit

OceanToken Setup

Security checks for vulnerabilities and agentic risk

Overview

This skill gives setup instructions for using OceanToken as an AI provider, including disclosed configuration changes and API key handling, without hidden code or deceptive behavior.

Install this only if you intend to route Codex, Claude Code, or your application through OceanToken. Review any proposed edits before approving them, prefer project-local git-ignored settings where possible, avoid committing API keys, and remember that prompts and API traffic will go to OceanToken when the provider profile is used.

A.I.G

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (8)

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
93% confidence
Finding

This section directs modification of ~/.codex/config.toml, a sensitive agent configuration file in the user's home directory. Even though it says to show a diff and ask permission, instructions targeting privileged config locations can persistently alter agent behavior, reroute future model traffic to a third-party endpoint, and create long-lived security impact beyond the current task.

Content

Scanner excerpt · SKILL.md (reported line 35)May include surrounding context.

md
## Codex on OceanToken

Add a provider and a profile to `~/.codex/config.toml` so the user's default
setup stays untouched and OceanToken is one flag away. Pick the model with
`search_models` (type `chat`, capability `tools`); OpenAI GPT-5 family models
are the safe choice for Codex.

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
97% confidence
Finding

This section instructs writing to ~/.claude/settings.json and includes storing ANTHROPIC_AUTH_TOKEN in plaintext in a settings file. That creates persistent secret exposure and changes the global provider endpoint for Claude Code, potentially redirecting all future requests and credentials to the third-party service if followed carelessly.

Content

Scanner excerpt · SKILL.md (reported line 60)May include surrounding context.

Claude Code talks the Anthropic Messages API, which OceanToken serves for its Claude models. For one project, put this in .claude/settings.local.json (git-ignored); for every project, in ~/.claude/settings.json:

json
{

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The activation description is overly broad and includes generic API setup, authentication, and error-troubleshooting requests. That broad scope can cause the skill to trigger in many normal support scenarios and steer the agent toward changing model/provider configuration or handling credentials when the user may not have explicitly asked for OceanToken, increasing the chance of inappropriate config edits or secret-handling actions.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 15)May include surrounding context.

md
# Setting up OceanToken as a provider

OceanToken's API is OpenAI-compatible at `https://api.oceantoken.ai/v1`
(chat completions, responses, images, videos, audio, embeddings), and Claude
models also answer the Anthropic Messages API at `https://api.oceantoken.ai`.
Auth is `Authorization: Bearer sk-...`.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 31)May include surrounding context.

md
# Setting up OceanToken as a provider

OceanToken's API is OpenAI-compatible at `https://api.oceantoken.ai/v1`
(chat completions, responses, images, videos, audio, embeddings), and Claude
models also answer the Anthropic Messages API at `https://api.oceantoken.ai`.
Auth is `Authorization: Bearer sk-...`.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 43)May include surrounding context.

md
# Setting up OceanToken as a provider

OceanToken's API is OpenAI-compatible at `https://api.oceantoken.ai/v1`
(chat completions, responses, images, videos, audio, embeddings), and Claude
models also answer the Anthropic Messages API at `https://api.oceantoken.ai`.
Auth is `Authorization: Bearer sk-...`.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 85)May include surrounding context.

md
# Setting up OceanToken as a provider

OceanToken's API is OpenAI-compatible at `https://api.oceantoken.ai/v1`
(chat completions, responses, images, videos, audio, embeddings), and Claude
models also answer the Anthropic Messages API at `https://api.oceantoken.ai`.
Auth is `Authorization: Bearer sk-...`.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill enables implicit invocation and uses broad wording that targets generic setup, API key, and error-resolution scenarios, which can cause the agent to activate this skill in situations the user did not clearly intend. That can redirect model-provider configuration, influence credentials handling, or trigger provider-specific guidance in unrelated contexts, creating a meaningful risk of unintended behavior.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.