
business-doc-evaluator
by danbrickey
AI Expert Team resources for Enterprise Data Platform modernization
SKILL.md
name: business-doc-evaluator description: Evaluates business documents (meeting notes, status reports, executive briefs, proposals, PRDs, technical docs, post-mortems, SOPs) against 9 quality principles to identify specific problems and provide actionable fixes. Use when the user explicitly asks to "evaluate", "review", or "assess" document quality, when something "feels off" about a document, or when checking if a document is ready to ship.
Business Document Quality Evaluator
Evaluates any business document against 9 principles of quality business writing, identifying specific failures and providing actionable fixes.
When to Use This Skill
Trigger this skill when the user:
- Explicitly asks to "evaluate", "review", or "assess" a document's quality
- Says something "feels off" or "isn't working" about a document
- Asks if a document is "ready to ship/send/share"
- Asks for "feedback on" a business document
- Mentions concerns about "AI-generated document" quality
- Asks "what's wrong with this" or "why isn't this document effective"
Do NOT trigger this skill for:
- Simple editing requests ("fix typos", "make this shorter", "rewrite this section")
- Creative writing (fiction, poetry, marketing copy)
- Code review (use different skills)
- Casual communication (Slack messages, informal emails)
- Documents still in active drafting (not ready for evaluation)
- Heavily domain-specific documents where you lack context (legal contracts, medical reports)
How This Works
This skill evaluates documents against 9 core principles (see references/principles.md for details). Each principle addresses a common failure mode in business writing. The evaluation produces:
- Principle-by-principle assessment - Which principles are violated and how
- Priority ranking - Most critical issues first
- Specific fixes - Not "make it better" but "do exactly this"
- Impact analysis - Why each issue matters
Evaluation Process
Step 1: Document Length Check
Before starting evaluation, check document length:
- Short documents (<1,000 words or <3 pages): Full evaluation appropriate
- Medium documents (1,000-2,500 words or 3-8 pages): Full evaluation appropriate
- Long documents (>2,500 words or >8 pages): Use streamlined approach (see guidelines below)
Step 2: Identify Document Type
Determine what type of document this is:
- Meeting notes
- Status report
- Executive brief / decision memo
- Project proposal
- PRD (Product Requirements Document)
- Technical documentation
- Post-mortem / incident report
- SOP (Standard Operating Procedure)
- Other business document
If the document type isn't obvious: Ask the user to clarify.
If the document serves multiple purposes (e.g., "meeting notes with status update"):
- Identify the primary purpose (what is the document mostly doing?)
- Apply the failure modes for that primary type
- Note in evaluation: "This document combines [type 1] and [type 2], evaluated primarily as [type 1]"
Step 3: Load Principles
CRITICAL: You MUST read the principles file before evaluation.
Use the view tool to read /mnt/skills/user/business-doc-evaluator/references/principles.md to load:
- All 9 principles with their tests
- Document-specific failure modes
- Prioritization hierarchy
Step 4: Quick Scan for Fundamental Issues
Before doing full evaluation, check for fundamental failures:
Principle 1 (Purpose) Check:
- Can you state in one sentence what this document is supposed to make someone do?
- Is there a clear decision or action requested?
Principle 6 (Input Quality) Check:
- Does the document contain specific facts, numbers, decisions, and examples?
- Or does it contain vague claims, estimated numbers, or inferred information?
If either Principle 1 or Principle 6 has a fundamental failure:
Consider stopping evaluation and returning:
OVERALL VERDICT: REJECT - Fundamental Issue
PRIMARY PROBLEM: [Principle 1 or 6 failure description]
EVIDENCE: [Specific examples from document]
RECOMMENDATION: Fix this core issue before detailed evaluation. The other principles don't matter until this is resolved.
NEXT STEPS: [What specifically needs to be gathered or clarified]
Only proceed with full evaluation if these two principles are at least PARTIAL (not complete FAIL).
Step 5: Evaluate Against All Principles
For each of the 9 principles, assess:
- PASS: Principle is satisfied
- FAIL: Principle is violated
- PARTIAL: Principle is partially satisfied with room for improvement
- N/A: Cannot assess (lack context, e.g., Principle 9 without workflow knowledge)
For each FAIL or PARTIAL, document:
- Which principle is violated
- Specific evidence from the document
- Impact of the violation
- Specific fix with exact location in document
Step 6: Structure Output
Use the output format specified in the "Output Format" section below.
Output Format
Standard Evaluation Output
DOCUMENT TYPE: [meeting notes / status report / etc]
OVERALL VERDICT: [SHIP / REVISE / REJECT]
- SHIP: Document is ready to use
- REVISE: Document has fixable issues that should be addressed
- REJECT: Document has fundamental problems (Principle 1 or 6 failures, wrong type, missing critical information)
DOCUMENT-SPECIFIC CHECKS:
[For the specific document type, check against known failure modes from principles.md]
Example for meeting notes:
✓ PASS: Decisions documented with decision-maker
✓ PASS: Action items have owner and deadline
✓ PASS: Date and attendees listed
✗ FAIL: No clear next steps identified
PRINCIPLE EVALUATION:
Principle 1 (Purpose): [PASS / PARTIAL / FAIL]
[If PARTIAL or FAIL: Specific problem, evidence, fix]
Principle 2 (Structure): [PASS / PARTIAL / FAIL]
[If PARTIAL or FAIL: Specific problem, evidence, fix]
Principle 3 (Constraints): [PASS / PARTIAL / FAIL]
[If PARTIAL or FAIL: Specific problem, evidence, fix]
Principle 4 (Self-Evaluation): [PASS / PARTIAL / FAIL]
[If PARTIAL or FAIL: Specific problem, evidence, fix]
Principle 5 (Failure Modes): [PASS / PARTIAL / FAIL]
[If PARTIAL or FAIL: Specific problem, evidence, fix]
Principle 6 (Input Quality): [PASS / PARTIAL / FAIL]
[If PARTIAL or FAIL: Specific problem, evidence, fix]
Principle 7 (Voice): [PASS / PARTIAL / FAIL / N/A]
[If PARTIAL or FAIL: Specific problem, evidence, fix]
Principle 8 (Iteration): [PASS / PARTIAL / FAIL / N/A]
[If PARTIAL or FAIL: Specific problem, evidence, fix]
Principle 9 (Workflows): [PASS / PARTIAL / FAIL / N/A]
[If PARTIAL or FAIL: Specific problem, evidence, fix]
[If N/A: "Cannot assess without organizational workflow context"]
PRIORITY FIXES:
[Maximum 5 fixes, prioritized using the hierarchy from principles.md]
1. [Highest priority - typically Tier 1: Principle 1 or 6]
Problem: [Specific issue]
Location: [Where in document]
Fix: [Exact action to take]
Why this matters: [Impact on document effectiveness]
2. [Second priority - typically Tier 2: Principle 3 or 2]
Problem: [Specific issue]
Location: [Where in document]
Fix: [Exact action to take]
Why this matters: [Impact]
3. [Third priority]
[Same format]
[Continue for up to 5 total fixes - don't overwhelm with everything]
WHAT WENT WELL: [Include only if OVERALL VERDICT is SHIP or REVISE with minor issues]
[2-3 specific things the document does right - be genuine, not patronizing]
Streamlined Output for Long Documents
For documents >2,500 words or >8 pages, use this streamlined approach:
DOCUMENT TYPE: [type]
DOCUMENT LENGTH: [word count / page count]
OVERALL VERDICT: [SHIP / REVISE / REJECT]
QUICK SCAN RESULTS:
- Principle 1 (Purpose): [assessment with 1 sentence explanation]
- Principle 6 (Input Quality): [assessment with 1 sentence explanation]
- Principle 3 (Constraints): [assessment with 1 sentence explanation]
TOP 3 PRIORITY FIXES:
[Only the 3 most critical issues with specific locations and fixes]
RECOMMENDATION:
[If document needs substantial work: "Recommend addressing priority fixes above, then requesting re-evaluation of specific sections rather than entire document."]
Evaluation Guidelines
Be specific, not generic:
- Bad: "This document needs more detail"
- Good: "Section 2 lacks metrics. Add: number of users affected, duration of impact, revenue loss"
Locate problems precisely:
- Bad: "The structure is unclear"
- Good: "Paragraphs 3-5 discuss solution before stating the problem. Move problem statement to paragraph 1"
Give actionable fixes:
- Bad: "Make this more concise"
- Good: "Cut paragraphs 4 and 7 (background context). Keep only paragraphs 1-3 and 8-10 (problem and solution)"
Prioritize using the tier system:
- Tier 1 issues (Principles 1, 6) always come first
- Within same tier, prioritize by impact on document function
- Maximum 5 fixes in output - more than that is overwhelming
- For documents with 10+ issues, group related fixes
Be honest but constructive:
- Call out real problems directly - "This document has no clear purpose" not "The purpose could be clearer"
- Acknowledge what works well (when verdict is SHIP or REVISE with minor issues)
- Remember the goal is improvement, not criticism
- Avoid hedge language in evaluation ("somewhat", "could be", "might benefit from")
Handle subjective principles carefully:
- Principle 7 (Voice): Look for generic AI phrases rather than judging authenticity
- Principle 9 (Workflows): Mark N/A if you lack organizational context
- When uncertain, explain your reasoning: "This appears to fail Principle X because [evidence], but this may be intentional for [reason]"
Common Evaluation Patterns
Pattern 1: No clear purpose (Principle 1 failure) This is the most common and most serious failure. If the document doesn't have a clear purpose, stop evaluation and return fundamental failure output. Don't evaluate structure, voice, etc. when there's no purpose to support.
Example: Meeting notes that transcribe everything said but don't capture decisions or action items.
Pattern 2: Built on assumptions (Principle 6 failure) AI filled gaps with plausible-sounding fiction. The document appears finished but contains invented data, inferred decisions, or estimated numbers. This requires stopping evaluation and identifying what actual information is missing.
Example: Project proposal with ROI numbers that weren't calculated, just estimated by AI.
Pattern 3: No constraints (Principle 3 failure) Document includes everything because nothing was explicitly excluded. This typically co-occurs with Principle 2 (Structure) failures.
Example: Executive brief that's 5 pages instead of 1 page, making it not actually a brief.
Pattern 4: Happy path only (Principle 5 failure) Technical docs, SOPs, and post-mortems especially: they show how things work when everything goes right, but not what to do when things break.
Example: Technical documentation with no troubleshooting section or error handling.
Pattern 5: Multiple fundamental failures When both Principle 1 and Principle 6 fail, the document is fundamentally broken. Use the REJECT verdict and recommend starting over with clear purpose and actual information.
Score
Total Score
Based on repository quality metrics
SKILL.mdファイルが含まれている
ライセンスが設定されている
100文字以上の説明がある
GitHub Stars 100以上
3ヶ月以内に更新がある
10回以上フォークされている
オープンIssueが50未満
プログラミング言語が設定されている
1つ以上のタグが設定されている
Reviews
Reviews coming soon