Security risk report for @forgecat/openai_skills_jupyter-notebook v0.0.11
Source Integrity
Low
Profile is sourced from OpenAI's official skills repository (https://github.com/openai/skills) with a specific commit hash and Apache-2.0 license.
Metadata is transparent: author, original repository, version, and commit are all documented and verifiable.
Agent Intent
Low
Content describes legitimate Jupyter notebook creation and editing workflows with no instructions to manipulate the AI, read credentials, exfiltrate data, or hide instructions.
References to helper scripts (new_notebook.py) and templates are standard development practices; no guidance poisoning or malicious knowledge injection is present.
Documentation is educational and procedural, focusing on notebook structure, patterns, and quality standards without any prompt injection, role hijacking, or system prompt leakage attempts.
Details
Evidence
Skill description: 'Use when the user asks to create, scaffold, or edit Jupyter notebooks'
Workflow step 2: 'Use the helper script to avoid hand-authoring raw notebook JSON'
No instructions to fetch external URLs, read ~/.ssh, ~/.aws, .env, or transmit system prompts
Permissions
Low
No tools or hooks are declared in the profile; the skill is purely instructional guidance for notebook creation.
References to local file operations (reading/writing .ipynb files) are scoped to the user's working directory and output folders, matching the stated purpose of notebook scaffolding and editing.
Environment setup (CODEX_HOME, JUPYTER_NOTEBOOK_CLI) is user-scoped and does not request elevated or excessive authority.
Details
Evidence
Declared tools: (none)
File operations are limited to tmp/jupyter-notebook/ and output/jupyter-notebook/ directories
No shell, file_delete, or high-risk tool categories are invoked
MCP Risk
Low
No MCP servers are defined in the profile.
The profile relies on local Python scripts (new_notebook.py) and standard library utilities, with no arbitrary binary execution or unrestricted network access.
Dependencies are optional and well-known (jupyterlab, ipykernel, uv) with no hidden or suspicious tool definitions.
ForgeCat