Back to skill

Security audit

AI Common Sense

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed AI-model reference skill with some caution-worthy permissions and API examples, but no hidden malicious behavior was found.

Install only if you want an AI-model reference helper that can search the web and read its reference files. Treat the included API examples as real external provider calls, use restricted development keys, avoid sensitive prompts in examples, and consider removing or limiting Bash permission if your runtime supports tighter tool scoping.

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

Note
Location
.
Finding

API Credentials May Be Exposed Through Command-Line Arguments and URLs

Content
View full analysis

Vulnerability Details

File Locations:

  • anthropic.md:50-61
  • cohere.md:39-47
  • deepseek.md:37-45
  • google.md:48-52
  • mistral.md:46-56
  • openai.md:70-80

Vulnerability Type: API credential exposure through process arguments and URL logging
Risk Level: Low

Vulnerable Code

anthropic.md:50-61

bash
curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-sonnet-4-6",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

cohere.md:39-47

bash
curl https://api.cohere.com/v2/chat \
  -H "Authorization: Bearer $COHERE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "command-a-reasoning",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

deepseek.md:37-45

bash
curl https://api.deepseek.com/chat/completions \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-chat",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

google.md:48-52

bash
curl "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent?key=$GOOGLE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"contents": [{"parts": [{"text": "Hello"}]}]}'

mistral.md:46-56

bash
curl https://api.mistral.ai/v1/chat/completions \
  -H "Authorization: Bearer $MISTRAL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mistral-small-latest",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

openai.md:70-80

bash
curl https://api.openai.com/v1/chat/completions \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt
...[truncated 2311 chars]
Remediation
View remediation

Remediation Suggestions

  1. Prefer provider SDK examples that read credentials directly from environment variables without placing their values in command-line arguments.
  2. For curl, load sensitive headers from a permission-restricted configuration file or protected file descriptor rather than an inline -H argument.
  3. Avoid query-string credentials wherever the provider supports a header-based authentication mechanism.
  4. Add an explicit warning that examples should be tested with restricted, disposable development keys rather than unrestricted production credentials.
  5. Recommend provider-side controls such as API restrictions, minimum required scopes, spending caps, rate limits, source restrictions, rotation, and expiration.
  6. Advise users to disable shell tracing and verify that command, proxy, and endpoint telemetry will not retain secrets.
  7. Redact authorization headers and credential-bearing URLs from examples, logs, screenshots, issue reports, and diagnostic output.

T05 · Unauthorized Access and Privilege Escalation

Note
Location
SKILL.md:1
Finding

Skill Grants Broad Shell and Filesystem Capabilities Beyond Its Core Reference Function

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:1-10
Vulnerability Type: Excessive tool permissions and violation of least privilege
Risk Level: Low

Vulnerable Code

yaml
---
name: ai-common-sense
description: |
  Use when user mentions model names, versions, pricing, API IDs,
  "which model should I use", "what's the latest model",
  "model comparison", "API pricing", "what models are available",
  "哪個模型最新", "模型比較", "API 定價",
  or when generating code that references specific AI model IDs.
version: 0.1.1
allowed-tools: Bash, Read, WebSearch, WebFetch
---

Technical Analysis

The Skill's declared purpose is to provide a model-information reference and verify stale model or pricing information. WebSearch and restricted web retrieval are reasonably connected to that purpose. General Bash and filesystem Read access provide materially broader capabilities than are required for model-reference lookup.

The only documented shell operations are package-manager commands used to inspect locally installed SDK versions. No existing instruction directs the agent to read sensitive files or execute destructive commands, and no malicious shell code was found. Nevertheless, granting general shell and filesystem capabilities increases the effect of any future prompt injection, compromised reference content, or agent-control failure.

Web-retrieved content can contain untrusted instructions. If an agent incorrectly follows such content while this Skill is active, the pre-authorized shell and read capabilities could be used outside the Skill's legitimate scope.

Attack Path

  1. The Skill is activated with general Bash, Read, WebSearch, and WebFetch access.
  2. The agent searches for current provider information or retrieves an external page.
  3. Search results or fetched content contain adversarial instructions, or the surrounding conversation supplies a malicious request.
  4. The age ...[truncated 1020 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove general Bash access unless local SDK-version inspection is essential to the declared functionality.
  2. Replace unrestricted shell execution with a narrowly scoped capability that can only retrieve approved package-version metadata.
  3. Remove general Read access if packaged reference files can be made available without arbitrary filesystem access.
  4. Restrict web access to official provider documentation and pricing domains where the Skill platform supports domain allowlisting.
  5. Require explicit user confirmation before any local command is run.
  6. Treat web pages and search results strictly as untrusted data, never as agent instructions.
  7. Run the Skill in a sandbox without access to user home directories, SSH material, cloud credentials, environment secrets, or unrelated project files.
  8. Document why every permitted tool is required and periodically review the allowlist for least-privilege compliance.
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 (14)

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 · CONTRIBUTING.md (reported line 53)May include surrounding context.

md
### For the Quick Reference in SKILL.md

If a model is important enough for the quick reference table (top ~15 models), also update `skills/ai-common-sense/SKILL.md`.

## Quality Standards

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The activation text includes broad conditions such as 'Use when user mentions model names, versions, pricing' and 'what models are available', which could match many ordinary conversations beyond the intended narrow skill context. The description does not provide explicit non-trigger examples or boundaries for when casual mention of models should not invoke the skill.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · anthropic.md (reported line 51)May include surrounding context.

API Quick Start

bash
curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \

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 · anthropic.md (reported line 51)May include surrounding context.

API Quick Start

bash
curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · cohere.md (reported line 40)May include surrounding context.

API Quick Start

bash
curl https://api.cohere.com/v2/chat \
  -H "Authorization: Bearer $COHERE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{

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 · cohere.md (reported line 40)May include surrounding context.

API Quick Start

bash
curl https://api.cohere.com/v2/chat \
  -H "Authorization: Bearer $COHERE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · deepseek.md (reported line 38)May include surrounding context.

API Quick Start

bash
curl https://api.deepseek.com/chat/completions \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{

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 · deepseek.md (reported line 38)May include surrounding context.

API Quick Start

bash
curl https://api.deepseek.com/chat/completions \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · google.md (reported line 49)May include surrounding context.

API Quick Start

bash
curl "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent?key=$GOOGLE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"contents": [{"parts": [{"text": "Hello"}]}]}'

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · mistral.md (reported line 47)May include surrounding context.

API Quick Start

bash
curl https://api.mistral.ai/v1/chat/completions \
  -H "Authorization: Bearer $MISTRAL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{

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 · mistral.md (reported line 47)May include surrounding context.

API Quick Start

bash
curl https://api.mistral.ai/v1/chat/completions \
  -H "Authorization: Bearer $MISTRAL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · openai.md (reported line 71)May include surrounding context.

API Quick Start

bash
curl https://api.openai.com/v1/chat/completions \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{

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 · openai.md (reported line 71)May include surrounding context.

API Quick Start

bash
curl https://api.openai.com/v1/chat/completions \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

This markdown file includes a curl example that sends user-supplied message content to the Mistral API using an API key, but the surrounding documentation does not mention that prompts are transmitted to a third-party service. For markdown files, externally transmitting user data should be accompanied by a clear warning when privacy or data handling could matter.

Content

No source excerpt is available for this finding.