Back to skill

Security audit

Company Skill Creator

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a coherent internal skill-creation helper, but its file-writing and packaging scripts have concrete safety gaps users should review before installing.

Install only if you are comfortable with a skill that reads internal docs you point it at and writes generated files. Use a dedicated output folder, review generated scripts before running them, do not paste live credentials or private keys, and avoid packaging skill folders that contain symlinks until the helper scripts are hardened.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (2)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/init_skill.py:47
Finding

Path Traversal and Unauthorized File Overwrite Through Unvalidated Skill Name

Content
View full analysis
Path: """Create a new skill directory with standard structure. Args: skill_name: Kebab-case skill name output_path: Parent directory where the skill folder will be created Returns: Path to the created skill directory """ skill_dir = output_path / skill_name skill_dir.mkdir(parents=True, exist_ok=True) # Create directories (skill_dir / "scripts").mkdir(exist_ok=True) (skill_dir / "references").mkdir(exist_ok=True) (skill_dir / "assets").mkdir(exist_ok=True) # Create SKILL.md skill_md = skill_dir / "SKILL.md" skill_md.write_text(SKILL_MD_TEMPLATE.format( skill_name=skill_name, skill_title=to_title(skill_name), )) # Create scripts/__init__.py (skill_dir / "scripts" / "__init__.py").write_text(INIT_PY_CONTENT) # Create .gitkeep files so empty dirs are tracked (skill_dir / "references" / ".gitkeep").write_text("") (skill_dir / "assets" / ".gitkeep").write_text("") return skill_dir ``` ```python skill_name = sys.argv[1] # Parse --path option output_path = Path.cwd() if "--path" in sys.argv: path_idx = sys.argv.index("--path") if path_idx + 1 < len(sys.argv): output_path = Path(sys.argv[path_idx + 1]) else: print("Error: --path requires a value") sys.exit(1) output_path = output_path.resolve() output_path.mkdir(parents=True, exist_ok=True) skill_dir = init_skill(skill_name, output_path) ``` ### Technical Analysis The command-line `skill_name` is treated as a filesystem path component without validating that it is a single kebab-case name. The expression `output_path / skill_name` does not guarantee containment within `output_path`: ...[truncated 1982 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/package_skill.py:91
Finding

Out-of-Root File Disclosure Through Symlinks During Skill Packaging

Content
View full analysis
/path/to/sensitive/config ``` 3. The victim runs `package_skill.py` against the crafted directory. 4. `rglob('*')` encounters the symlink. 5. `is_file()` follows ...[truncated 1044 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (11)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared description presents a high-level interactive skill-building assistant for internal company workflows, including requirement elicitation and automatic generation of missing artifacts. The supplied code does not implement any such interactive workflow or content generation. It simply creates directories and placeholder files on disk for a new skill skeleton based on a provided name and optional output path. While scaffolding a skill directory is loosely related to skill creation, the actual behavior is materially narrower and different from the declared purpose, so this is a clear mismatch.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description promises an interactive internal-skill creation assistant that helps collect requirements and generate missing artifacts for company workflows and tools. The supplied code does none of that. It only checks that a given skill directory exists, requires SKILL.md, runs a validator, and zips the folder into a .skill package while excluding some files. This is a materially different primary purpose: packaging an already-created skill rather than creating one interactively. No evidence of user guidance, document/spec generation, workflow automation, or internal API/process integration appears in the code.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description presents a user-facing interactive assistant for creating company-internal skills and collecting or generating needed materials. The supplied code does not implement any interactive creation flow, requirement gathering, document generation, or company workflow automation. Instead, it performs static validation of an existing skill's SKILL.md metadata and exits with a success/failure status. This is a materially different primary purpose, so the description does not accurately represent the code.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

There is a substantial mismatch between the declared purpose and the supplied code chunk. The description promises an interactive internal-skill creation assistant with guidance and generation features, but the code only reads a local SKILL.md file and extracts metadata from frontmatter. This parsing utility could be a supporting component of such a system, but on its own it does not implement the declared primary purpose. Therefore the code chunk does not accurately represent the described behavior.

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill description says to use this skill whenever the user mentions creating a company skill, internal skill, work skill, or automating a company workflow, which is intentionally broad. Over-broad trigger conditions can cause the skill to activate in unrelated enterprise conversations, pulling in file access, generation, and packaging behaviors where they were not intended and increasing the chance of accidental handling of sensitive internal material.

Content

No source excerpt is available for this finding.

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · SKILL.md (reported line 161)May include surrounding context.

md
---

## Phase 3: Write SKILL.md

Now compose the SKILL.md. This is the heart of the skill — follow the same patterns as skill-creator but tuned for company skills.

Vague Triggers

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The authoring guidance explicitly tells generated skills to be 'pushy' and to err on the side of triggering. That propagates an unsafe design pattern into downstream skills, making accidental activation systemic and potentially broadening access to internal docs, schemas, and scripts in contexts where the user did not clearly request such processing.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
81% confidence
Finding

The skill instructs the agent to read user-provided files, fetch referenced materials, scaffold directories, generate files, and package artifacts, but it does not declare any explicit tool scope or permissions boundaries. That mismatch increases the risk of unintended file access or modification because the runtime capabilities needed for the workflow are broader than what the metadata communicates or constrains.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The instruction mandates marking guessed parts with the Chinese phrase '⚠️ 待确认' rather than allowing the user's preferred language. This is a locale/language constraint that is not presented as optional and is not justified as region-specific.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The file instructs generated reference files and confidence markings to use '⚠️ 待确认' in multiple places. Because the skill does not offer language choice or explain why Chinese is required, this violates the language/locale policy for natural-language content.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The agent explicitly asks users for authentication-related materials such as an 'Authentication credentials setup guide' and API artifacts that often contain secrets, but it provides no warning to avoid sharing live credentials, tokens, connection strings, or internal-only sensitive configuration. In an interactive skill-creation context, users may paste secrets directly into the chat, creating unnecessary exposure of credentials and sensitive system details to the AI workflow.

Content

No source excerpt is available for this finding.