Research Pipeline Planner

Quick start

1. Inspect the current project, harness state, and research-commitment.json before asking broad reset questions.
2. Use this skill as the default first pass for project-level sequencing, persistent coordination, or work spanning two or more research stages.
3. Stay standalone for a direct one-stage task; use orchestrated mode when work must survive context loss, retries, sessions, tools, or agent handoffs.
4. In orchestrated mode, initialize the suite with scripts/init_research_pack.py, then use scripts/harness_runtime.py as the only writer for work-item and run state.
5. Freeze one paper identity before moving from exploration into a paper-bearing route; keep one active paper identity and one active work item by default.
6. Freeze each work item with explicit inputs, outputs, acceptance checks, permissions, budgets, dependencies, predecessor failures, evidence class, and relation to the commitment contract.
7. For experiment execution, bind the work item to one frozen claim map, run block, decision gate, and declared snapshot set using references/experiment-execution-binding.md.
8. Run the assigned stage skill as a bounded worker, require an episode package, and authorize transition only after verifier inspection.
9. Apply references/epistemic-assurance-contract.md, references/agentic-harness-contract.md, references/stage-gates.md, and references/research-commitment-contract.md as appropriate.
10. Treat D3-D4 identity changes as pivot requests, not ordinary revisions.

Routing triggers

Prefer this skill before a sibling stage skill when the user asks for a research roadmap, research agenda, project plan, phase plan, milestones, what to do next, repo organization, notes-to-plan conversion, or coordination across literature review, idea discovery, novelty review, experiments, results audit, paper planning, review, and rebuttal.