What is the QA threshold?

The default QA gate threshold is 70; sections scoring below 70 are flagged for refinement.

Can QA run automatically after report generation?

Yes. Automatic mode (via QAGateHook) runs after generation and logs warnings with feedback if the score is below 70.

How many refinement cycles are recommended?

Max 2 refinement cycles are recommended; always require user approval before re-research.

qa-iteration

Scanned

npx machina-cli add skill blisspixel/primr/qa-iteration --openclaw

Files (1)

SKILL.md

2.9 KB

QA Iteration

Primr grades reports on a 0-100 scale across four dimensions. See references/scoring.md for the scoring framework and common issues.

Operational Capabilities

Run Quality Assessment

run_qa(company="Acme Corp")
run_qa(file_path="output/acme_corp/report.docx")
run_qa(company="Acme Corp", detailed=True)

Interpret Results

The result includes an overall score, per-dimension scores (clarity, completeness, insight depth, accuracy), and per-section feedback with specific issues.

Section Refinement

For sections scoring below 70:

Identify specific issues from feedback
Determine if additional research is needed
Either request targeted deep research or suggest manual edits

Refinement Workflow

Automatic (via QAGateHook): Runs after report generation. If score < 70, logs a warning and includes feedback in the result. User decides on action.

Manual (user-triggered):

User requests QA: run_qa(company)
Review scores and feedback
For low-scoring sections, identify root cause and suggest approach
Execute refinement if approved

Section Priorities

When refining, address sections in this order:

Executive Summary -- sets tone for entire report
Financial Analysis -- factual accuracy is critical
Competitive Landscape -- highest strategic value
SWOT Analysis -- synthesis quality matters

Example Workflow

User: "Check the quality of the Acme Corp report"

1. run_qa("Acme Corp")

   Results:
   Overall: 72 (B)
   - Clarity: 85
   - Completeness: 65
   - Insight Depth: 75
   - Accuracy: 63

2. Present findings:
   "The Acme Corp report scores 72/100 (B grade).

   Strong: Clarity (85), Insight Depth (75)

   Needs work:
   - Completeness (65): Missing competitor TechCorp
   - Accuracy (63): Revenue figures need citations

   Would you like me to:
   1. Run targeted research on TechCorp?
   2. Find citations for financial claims?
   3. Both?"

3. If user approves research:
   research_company(company="TechCorp", url="https://techcorp.com", mode="scrape")

4. After additional research:
   "I've gathered information on TechCorp.
    The report can now be updated with:
    - TechCorp's market position
    - Competitive comparison

    Shall I regenerate the Competitive Landscape section?"

Constraints

Threshold: Default QA gate threshold is 70
Iteration Limit: Max 2 refinement cycles recommended
Cost Awareness: Each refinement may incur API costs
User Approval: Always get approval before re-research

Source

git clone https://github.com/blisspixel/primr/blob/main/skills/qa-iteration/SKILL.mdView on GitHub

Overview

QA Iteration grades research reports on a 0-100 scale across four dimensions: clarity, completeness, insight depth, and accuracy. It provides an overall score, per-dimension scores, and per-section feedback to guide targeted refinements, especially when sections fall below the 70 threshold.

How This Skill Works

Use run_qa to evaluate a report file or a company. The tool returns an overall score, per-dimension scores, and detailed section feedback. For sections scoring below 70, it identifies issues and suggests targeted deep research or manual edits; refinements can be automatic after generation via QAGateHook or manual per user request, following the priority order (Executive Summary, Financial Analysis, Competitive Landscape, SWOT).

When to Use It

User asks to check, grade, review, or improve a research report.
You need to verify report quality after generation before submission.
A section scores below 70 and requires targeted refinement.
You want to decide whether to perform targeted deep research or manual edits.
Planning refinement cycles with user approval and awareness of potential API costs.

Quick Start

Step 1: Run QA on the report: run_qa(company="Acme Corp") or run_qa(file_path="output/acme_corp/report.docx")
Step 2: Review the results and identify sections scoring below 70 with concrete issues
Step 3: With user approval, perform targeted research or manual edits and regenerate affected sections

Best Practices

Run with detailed per-section results by using detailed=True to surface issues.
Address sections in priority order: Executive Summary, Financial Analysis, Competitive Landscape, SWOT.
For low-scoring sections, identify root causes from feedback before taking action.
Obtain explicit user approval before performing re-research or edits.
Limit refinement cycles to two, and monitor potential costs.

Example Use Cases

User: 'Check the quality of the Acme Corp report' → run_qa("Acme Corp"), results show Overall 72; discuss actionable next steps and request targeted research for missing data in Competitive Landscape.
Automatic QA gate: After report generation, run_qa triggers a warning if score < 70 and surfaces detailed feedback for reviewer decision.
Targeted research: For a missing competitor data point in Financial Analysis, call research_company and then insert citations and figures after approval.
Section regeneration: After adding TechCorp data, regenerate the Competitive Landscape section to reflect new insights and improved accuracy.
Approval-driven refinement: User approves a second refinement cycle only after reviewing root causes and expected impact.

Frequently Asked Questions

Add this skill to your agents