eda-research-chain

Run a full EDA research chain from knowledge exploration and paper triage to idea debate, hypothesis experiment design, implementation, git versioning, validation, and retrospective.

EDA Research Chain

When to use

Use this skill when the user asks for an end-to-end research workflow, not a single isolated experiment step.

Scope

This skill orchestrates the full chain and delegates each stage to specialized skills. Global governance still follows AGENTS.md.

Stage flow

  1. Bootstrap chain workspace:
python3 scripts/common/init_research_chain.py --tag <tag>
  1. Knowledge exploration:
  • use eda-knowledge-explorer to map local knowledge gaps,
  • produce 01_knowledge/knowledge_gap_map.md.
  1. Literature retrieval + local parsing:
  • use eda-paper-fetch to generate download queue,
  • use eda-pdf-local-summary after local PDFs are available,
  • produce 02_literature/paper_download_queue.tsv and 02_literature/local_paper_summary_index.md.
  1. Idea brainstorming and debate:
  • use eda-idea-debate-lab,
  • produce 03_idea_debate/idea_brainstorm.md and 03_idea_debate/pro_con_debate.md.
  1. Hypothesis -> experiment design:
  • use eda-hypothesis-experiment-designer,
  • run eda-preflight-reflect before expensive submissions,
  • produce 04_hypothesis_design/hypothesis_experiment_matrix.tsv.
  1. Method implementation:
  • use eda-method-implementer,
  • produce 05_implementation/implementation_plan.md.
  1. Multi-version development and integration:
  • use git-version-control,
  • produce 06_versioning/version_plan.md and version delta notes.
  1. Validation and decision:
  • execute via eda-loop,
  • use validation tools/skills (delay-model-gate-evaluator, execution contract),
  • produce 07_validation/validation_summary.md.
  1. Retrospective:
  • use eda-retro,
  • produce 08_retro/research_retro.md.
  1. Version milestone summary slides (conditional):
  • do not generate slides for every process stage by default.
  • generate a PDF summary only when a new version achieves a validated milestone improvement (e.g., first reach >=2% power reduction with non-worse area/timing, then higher milestones).
  • summary should include: method delta, key conclusion, evidence table, and next-step plan.
  1. Guard completeness:
python3 scripts/common/research_chain_guard.py --chain-dir <chain_dir> --out-prefix <prefix>

Hard rules

  1. Do not skip the hypothesis design stage before implementation.
  2. Do not promote a method without explicit validation artifact.
  3. Keep each stage artifact path explicit and auditable.
  4. If critical guard checks fail, block chain completion.

Reference

Load when needed:

  1. references/chain-checklist.md