ds-review
This skill should be used when running Phase 4 of the /ds workflow or reviewing data analysis methodology.
npx skills add majiayu000/claude-skill-registry --skill ds-review-edwinhu-workflows-3 --agent claude-code
Same command for any agent — swap --agent for codex, cursor, copilot.
Weekly change comes from our own snapshots, not the repository page — it measures attention, not adoption.
What it does
The skill is invoked to oversee Phase 4 of the /ds workflow or to review data analysis methodology. It sets up a review framework, including a post-tool guard, pre-tool checks, and read-only reviewer prompts. It provides a decision point to choose between a single reviewer or a parallel, three-reviewer model, and outlines the workflow for initiating, monitoring, and reconciling reviews, including a three-pass reconciliation protocol and an audit loop with a defined maximum number of cycles. It enforces reporting of issues only if confidence is at least 80% and directs flow to either approval or changes required, with a process to load the ds-verify skill upon approval. The skill also includes policy for context monitoring, review strategy prompts, and constraints loading, and uses tools such as CLAUDE_CODE and various Python hook scripts for guard checks.
How it works
- Hooks are defined for PreToolUse and PostToolUse to run Python guard scripts around agent actions.
- When invoked, announce with a Phase 4 context and present a Context Monitoring table to guide action based on remaining context.
- Offer a Review Strategy Choice: either Single reviewer or Parallel review. If Single reviewer is chosen, proceed to The Iron Law of DS Review; if Parallel is chosen, enable Parallel Review (Research-Grade).
- In Parallel Review, ensure prereqs: SPEC.md, PLAN.md, LEARNINGS.md exist and Analysis files identified. Determine applicability by the use-case rules, then create a Team and spawn three reviewers with read-only tools.
- Provide reviewer prompts with variables substituted from the analysis context (ANALYSIS_FILES, SPEC_CONTEXT, PLAN_TASKS, LEARNINGS_PIPELINE, PLUGIN_ROOT) and link each reviewer to their focus area: Methodology, Reproducibility, Code Quality.
- Lead monitoring and reconciliation occur through a three-pass protocol: Deduplication, Prioritization, and Integration Check, followed by a Final Verdict (APPROVED or CHANGES REQUIRED).
- If APPROVED, load the ds-verify skill; if CHANGES REQUIRED, return to /ds-implement. Enforce a maximum of 3 review cycles and escalate if unresolved.
- The Iron Law sections restrict reporting to issues with >= 80% confidence and require re-review after fixes per the Re-Review rules.
When to use it
Use this skill when running Phase 4 of the /ds workflow or when reviewing data analysis methodology, particularly when a structured, multi-review process is appropriate or when ensuring an audit trail and reproducibility standards are needed.
What it can touch
- Tools: CLAUDE_CODE is declared as a tool. Reviewers are constrained to Read, Glob, Grep, and Bash(read-only). Review prompts substitute variables from the analysis context and use provided plugin root paths for guard scripts.
Caveats
- The skill enforces the Iron Law of DS Review: report issues only with confidence >= 80%.
- It relies on protected workflows and guard scripts to prevent main-chat-code leakage and ensures read-only reviewer prompts.
- If prerequisites or constraints are unavailable, flows may default to single-reviewer mode or halt and return to /ds-implement.
Announce: "Using ds-review (Phase 4) to check methodology and quality." ## Context Monitoring | Level | Remaining Context | Action | |-------|------------------|--------| | Normal | >35% | Proceed normally | | Warning | 25-35% | Complete current review cycle, then trigger ds-handoff | | Critical | ≤25% | Immediately trigger ds-handoff — do not start new review cycles | ## Review Strategy Choice After announcing phase, choose review strategy. **Skip this choice when:** - Exploratory analysis (one-off, not for publication) - Trivial changes (formatting, documentation) - Internal reporting (low-stakes, quick turnaround) - Single notebook with < 100 LOC **Otherwise, ask the user:** ```python AskUserQuestion(questions=[{ "question": "How should we review this analysis?", "header": "Review Strategy", "options": [ {"label": "Single reviewer (Default)", "description": "Combined review covering methodology, data quality, and reproducibility. Faster, lower overhead."}, {"label": "Parallel review (Research-grade)", "description": "Spawn 3 specialized reviewers (Methodology, Reproducibility, Code quality). Use for publications, high-stakes decisions, or research-grade work. Requires reconcilia
- Context Monitoring
- Review Strategy Choice
- Parallel Review (Research-Grade)
- 1. Prerequisites Check
- 2. When to Use Parallel Review
- 3. Create Team and Spawn Reviewers
- 4. Lead Monitoring
- 5. Reconciliation Protocol (3 Passes)
- 6. Final Verdict
- The Iron Law of DS Review
- The Iron Law of Re-Review
- The Audit-Fix Loop (Max 3 Iterations)
- Rationalization Prevention (Re-Review)
- Why Skipping Re-Review Hurts the Thing You Care About Most
What does the ds-review skill do?
This skill should be used when running Phase 4 of the /ds workflow or reviewing data analysis methodology.
How do I install it?
Run `npx skills add majiayu000/claude-skill-registry --skill ds-review-edwinhu-workflows-3 --agent claude-code` — it drops the skill into your project so the agent can pick it up. Swap the --agent value for codex, cursor or copilot if you use one of those.
Where does this skill come from?
From majiayu000/claude-skill-registry, a repository with 534 stars. We read it straight from the repository tree rather than a submitted listing, so what you see here is what is actually published.
Is a popular skill a good skill?
Not necessarily. Stars measure attention, not adoption — a repository can trend for a week and be abandoned. That is why we show the weekly change from our own snapshots next to the total, instead of a single flattering number.