ce-doc-review

Reviews requirements and plan documents using parallel persona agents with tiered finding resolution.

Updated Mar 17, 2026
One-click install
npx skills add https://github.com/Norfolk-Group/marcela-norfolk-ai --skill ce-doc-review-norfolk-group
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: ce-doc-review
Source: https://github.com/Norfolk-Group/marcela-norfolk-ai/tree/main/skills/compound-engineering/skills/ce-doc-review
Command: npx skills add https://github.com/Norfolk-Group/marcela-norfolk-ai --skill ce-doc-review-norfolk-group

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Requirements and plan documents often ship with contradictions, omissions, and scope issues that only surface during implementation. This Skill reviews such documents through multiple specialized reviewer personas running in parallel, then routes findings through a structured decision workflow so issues are caught and resolved before work begins. ## Core Features & Use Cases - Multi-Persona Parallel Review: Dispatches coherence and feasibility reviewers plus conditional personas (product-lens, design-lens, security-lens, scope-guardian, adversarial) selected based on document content signals. - Tiered Finding Resolution: Classifies findings as safe_auto (applied silently), gated_auto (confirmed fixes), manual (user judgment), or FYI observations, with anchored confidence scores and deduplication across personas. - Interactive and Headless Modes: Supports an interactive walk-through with per-finding decisions and bulk-action previews, or a headless mode returning structured findings for automated callers. - Use Case: After drafting an implementation plan in docs/plans/, run the review to have security, scope, and feasibility personas flag issues like missing rollback strategies or unverifiable premises, then apply safe fixes and defer open questions directly into the document. ## Quick Start Ask the agent to review your plan document, for example: run ce-doc-review on docs/plans/my-feature-plan.md and walk me through the findings.

Frequently Asked Questions about ce-doc-review

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I review a plan document with multiple AI reviewer personas?

Invoke the skill with the document path, and it dispatches coherence and feasibility reviewers plus conditional personas like security-lens or scope-guardian in parallel. Findings are synthesized, deduplicated, and presented grouped by severity for your decision.

What is the difference between interactive and headless mode in document review?

Interactive mode walks you through each finding with Apply, Defer, or Skip options and bulk-action previews. Headless mode, triggered by the mode:headless argument, applies safe fixes silently and returns all other findings as structured text without prompting.

How are document review findings classified and routed?

Findings carry an autofix class: safe_auto fixes apply silently at confidence 100, gated_auto fixes require one-click confirmation, and manual findings need user judgment. Confidence anchors 0 and 25 are dropped, 50 routes to an FYI subsection, and 75 or 100 enter the actionable tier.

Can deferred review findings be tracked inside the document itself?

Yes. Choosing Defer appends the finding to a Deferred / Open Questions section under a dated subsection, with an HTML-comment dedup key preventing duplicate entries across retries. Cross-session persistence is out of scope; each new invocation starts fresh.

What happens when reviewer personas disagree on a document issue?

Contradictory findings are merged into a single combined finding classified as manual and framed as a tradeoff rather than a verdict. A deterministic tie-break picks the recommended action in the order Skip, then Defer, then Apply, so results stay reproducible.

When should I not use multi-persona document review?

It is designed for requirements and plan markdown documents, not code review or style linting. Pedantic style nitpicks, linter-catchable errors, and pre-existing issues the document did not introduce are actively suppressed as non-findings.