deepdive
Deepdive skill for Claude Code — 12-phase research pipeline: plan-review gate, parallel sub-agent search, claims-ledger triangulation with dissent protection, r
What is deepdive?
Deepdive skill for Claude Code — 12-phase research pipeline: plan-review gate, parallel sub-agent search, claims-ledger triangulation with dissent protection, relevance × authority evidence filter, multi-angle red team, four-layer citation verification. 105 blocks, 29 channels, 460+ stat sources, 47 APIs, 1072 verified endpoints.
Overview of deepdive
What it is
Deepdive is a structured meta-research skill for Claude Code. It turns a research request into a 13-phase pipeline with hypothesis testing, parallel sub-agent search, source triangulation, and adversarial review. The output is a folder containing a plan document, source files with verbatim quotes, and a final report. Every claim traces to a specific source file.
Key features
- 13-phase research pipeline including reframing, genre selection, plan creation, capability discovery, plan-review gate, search, claims-ledger triangulation, evidence filter, wiki reconcile, synthesis with multi-angle red team, verification, refresh targets, and decision walkthrough.
- Parallel sub-agent search across 29 named channels and 460+ stat sources.
- Claims-ledger triangulation with dissent protection and relevance × authority evidence filter.
- Multi-angle red team and four-layer citation verification including faithfulness checks (claim⊨quote).
- Plan-review gate that shows reframing, hypotheses, genre, and channels for approval before execution.
- Input-level skepticism markers (vendor, self-reported, disputed) on sources before synthesis.
- Validation script (validate_phases.py) that checks mandatory phase artifacts exist.
Use cases
- Investigating trade-offs between technologies (e.g., Postgres logical replication vs. CDC tooling).
- Validating hypotheses with documented evidence.
- Conducting structured research that produces reusable findings and traceable sources.
- Decision-making with adversarial review and decision walkthrough.
Getting started
bash git clone https://github.com/Socialpranker/deepdive.git ~/.claude/skills/deepdive For Claude Desktop, clone the repository, zip as .skill bundle, and upload via Claude.app → Settings → Skills → Add Skill. For other LLMs, load SKILL.md and references/*.md into context manually.
License and pricing
Licensed under MIT. Open source.
Based on the project's public README and repository metadata.