Ai Paper Pipeline

Build or improve a top-tier AI conference paper workflow for NeurIPS, ICML, ICLR, and similar venues. Use when the user asks to generate a paper pipeline, or...

Security Scan

The skill's stated purpose (scaffolding an AI-paper workflow) is plausible, but its runtime instructions expect access to many local and external resources (API keys, hardware detection, file-system paths, code execution) that the registry metadata does not declare — this mismatch is concerning and needs clarification before use.

Purpose & Capability

The skill claims to scaffold/packaging paper projects and initialize experiment workflows (reasonable). However the included instructions and MEGA_PROMPT explicitly expect networked model APIs, literature APIs, Kaggle/Tavily tokens, and access to local project source paths and hardware detection. The registry shows no required env vars, binaries, or config paths — that is inconsistent (the skill likely needs credentials and runtime access beyond what it declares).

Instruction Scope

SKILL.md and MEGA_PROMPT instruct the agent to: read and create many project files, detect local GPU/CPU (hardware detection), call external literature APIs (OpenAlex, Semantic Scholar, arXiv, Google Scholar), call large-model endpoints (OPENAI_API_BASE/KEY, OPENAI_MODEL_NAME), use Kaggle/Tavily tokens, run sandboxed experiments, install libraries, and commit changes. Those are broad actions touching local filesystem, environment variables, network endpoints, and code execution — all legitimate for a research pipeline but they are not limited or reflected in the declared requirements and therefore grant wide runtime scope.

Install Mechanism

No install spec and no packaged code — this is instruction-only, so the skill itself does not download or execute third-party installers. That reduces supply-chain risk. Note: the instructions do tell the agent it may install Python libraries in the environment at runtime (venv/docker), which is normal for experiments but happens outside the skill packaging.

Credentials

Registry metadata declares no required environment variables or credentials, but MEGA_PROMPT lists several sensitive env vars (OPENAI_API_BASE, OPENAI_API_KEY, OPENAI_MODEL_NAME, KAGGLE_API_TOKEN, TAVILY_API_KEY) and expects access to local project paths and possibly system-level hardware info. This mismatch is disproportionate and could result in the agent attempting to read secrets or ask for credentials not declared up-front.

Persistence & Privilege

The skill is not always-enabled and does not request elevated persistent privileges in metadata. It instructs the agent to create and commit files in the workspace (normal for a scaffolding skill). No evidence it attempts to modify other skills or global agent settings.

What to consider before installing

This skill is a heavy-duty instruction-only workflow for building paper projects and running experiments. Before installing or invoking it: 1) Confirm which credentials and services you allow it to access — the files reference OPENAI_API_KEY, Kaggle and other tokens but the registry does not declare them. 2) Only run it in an isolated/dev environment if you expect it to run code, detect hardware, or install packages — it can run experiments that consume compute and access local paths. 3) If you plan to let it use model APIs, supply ephemeral keys or restrict scopes; never expose long-lived admin keys. 4) Review the MEGA_PROMPT.md and RESTRICTS templates line-by-line for any mentions of private paths, network endpoints, or data sources you don't want the agent to touch. 5) Require explicit human approval for any step that executes experiments, installs packages, or commits/pushes to remote repos. If the author can clarify and update registry metadata to list the required env vars and permissions, the coherence concerns would be resolved.

Turn a rough paper idea or a long "mega prompt" into a reusable, reality-grounded paper project scaffold.

What this skill should do

• Normalize a user's long paper-workflow prompt into a maintainable skill/project structure.
• Keep the main workflow concise in SKILL.md and push bulky reference text into references/.
• Preserve academic-integrity constraints: no fabricated experiments, no fake citations, no unsupported claims.
• Prefer creating reusable project scaffolding over dumping one giant prompt blob.

Default workflow

1. Identify whether the user wants one of these:
   • skill cleanup / packaging for the paper workflow itself
   • project initialization for a specific paper
   • template ingestion from a pasted mega prompt
2. If the user pasted a large workflow prompt, extract and organize it into:
   • SKILL.md for concise usage instructions
   • references/ for long-form reference content
   • templates/ for starter files like RESTRICTS.example.yaml
3. Keep only trigger logic, workflow guidance, and file navigation in SKILL.md.
4. Put long source material, detailed prompts, and heavy policy text in references/.
5. If the user wants a paper project initialized, create at minimum:
   • MEGA_PROMPT.md
   • RESTRICTS.yaml
   • PROGRESS.md
   • plans/
   • code/, data/, docs/, results/
   • paper/mypaper/main.tex
   • paper/mypaper/sections/
6. After edits, package or commit changes if appropriate.

File layout for this skill

ai-paper-pipeline/
├── SKILL.md
├── MEGA_PROMPT.md
├── references/
│   ├── full-pipeline-template.md
│   └── project-scaffold.md
└── templates/
    └── RESTRICTS.example.yaml

When to read extra files

• Read MEGA_PROMPT.md when you need the concise built-in version of the 25-stage workflow.
• Read references/full-pipeline-template.md when the user wants the verbose original template or asks to reconstruct/port the full prompt.
• Read references/project-scaffold.md when the user wants to initialize a concrete paper project directory.
• Read templates/RESTRICTS.example.yaml when initializing a new paper project or drafting a restrictions file.

Working rules

• Treat the paper as a real research artifact, not a vibe-writing exercise.
• Never claim experiments, datasets, ablations, or statistical tests that are not actually present.
• Never keep huge duplicated prompt text in multiple files.
• Prefer editable project artifacts over giant single-message outputs.
• Keep the paper workflow cyclical: literature → design → run → analyze → draft → review → revise.

Good outputs

A. User says: "整理成一个 Skill"

Do this:

• Clean up the current skill folder.
• Convert ad-hoc text into proper SKILL.md + references/ + templates/.
• Keep SKILL.md concise and reusable.

B. User says: "按这个模板起一个论文项目"

Do this:

• Create a new <project>-paper/ scaffold.
• Copy in starter files.
• Replace placeholders with project-specific metadata where provided.

C. User says: "把这份 mega prompt 落库"

Do this:

• Save the raw template in references/ or project root.
• Avoid bloating SKILL.md with the full raw text.

Final step

After modifying this skill or creating paper-project files in the workspace, commit the changes with a clear git message.