| name | paperclip-pipeline |
| description | Coordinate multi-agent academic research workflows with sanitized Paperclip-style handoffs, verification gates, and release-safe outputs. |
Paperclip Pipeline
Use this skill when a research task spans multiple AI Research Toolkit phases and needs multiple agents, explicit handoff contracts, verification evidence, and release-safe output handling.
When to Use
Use this skill for:
- Literature review projects that require search, PDF processing, analysis, writing, and knowledge-base updates.
- Multi-agent research tasks where work must be split across independent reviewers, writers, auditors, or curators.
- Release preparation tasks that must separate local debug work from public, sanitized GitHub output.
- Long-running research workflows that need checkpointing, handoff packets, and evidence-based acceptance.
Do not use this skill for:
- A single-paper quick summary.
- A one-file editing task.
- Directly connecting to a private Paperclip deployment.
- Publishing private runtime state, tokens, local paths, or account details.
Inputs
Before starting, collect:
- Research goal.
- Target phases from
01-discovery through 06-presentation.
- Source material location or public links.
- Privacy boundary: public, private, and excluded outputs.
- Required deliverables.
- Verification commands or acceptance evidence.
Workflow
- Map the request to phases.
- Split the work into independent agent tasks.
- Assign each task a concrete output path and acceptance criteria.
- Require each agent to return a handoff packet.
- Verify key files and command outputs directly.
- Synthesize final research artifacts.
- Run the security boundary check before producing public outputs.
- Prepare release-safe summaries, docs, or module changes.
Handoff Contract
Every handoff must include:
- Task objective.
- Files read.
- Files modified.
- Commands run.
- Evidence produced.
- Assumptions made.
- Open risks.
- Next recommended step.
If a handoff omits evidence, treat it as incomplete.
Verification Rule
Do not trust agent summaries blindly. Verify the final file tree, relevant diffs, and critical command outputs before marking a task complete.
Security Boundary
Public outputs must not include:
- API keys, tokens, cookies, passwords, private keys, or secrets.
- Real service URLs, internal URLs, IP addresses, ports, database strings, OAuth values, team IDs, workspace IDs, or job IDs.
- Local absolute paths, user home directories, private Obsidian vault paths, private Zotero paths, or private Paperclip paths.
- Runtime logs, cache files, coverage artifacts, local task records, or private deployment configs.
Use placeholders such as <PAPERCLIP_API_URL> and <PUBLIC_WORKSPACE_NAME> in public templates.
Phase Integration
Paperclip Pipeline coordinates the existing toolkit phases:
| Phase | Role in pipeline |
|---|
| 01 Discovery | Find papers, collect metadata, prepare Zotero intake. |
| 02 Processing | Convert PDFs and documents into Markdown. |
| 03 Analysis | Review papers, synthesize evidence, audit citations. |
| 04 Writing | Draft sections, polish prose, prepare responses or manuscripts. |
| 05 Knowledge | Ingest notes, update graph views, maintain research memory. |
| 06 Presentation | Produce slides, figures, diagrams, and meeting artifacts. |
| 07 Pipeline | Decompose tasks, coordinate agents, verify handoffs, and prepare release-safe outputs. |
References
references/workflow.md — end-to-end orchestration workflow.references/handoff-contract.md — required handoff packet schema.references/security-boundary.md — public/private output boundary.references/release-sync.md — debug-to-release synchronization protocol.
