AI-Assisted Report Revision Workflow

When to Use This Skill

• User wants to revise or enhance a performance test report with AI-generated content
• User mentions report revision, AI summary, executive summary, key observations, or issues table
• User wants to iteratively refine report sections with feedback
• User wants to publish a revised report to Confluence
• This workflow supports both single_run reports and comparison reports

Reference

This section provides context for humans and capable models. For the step-by-step execution instructions, skip to the Execution section below.

What This Workflow Does

Orchestrates the AI-assisted revision of performance test reports using a Human-In-The-Loop (HITL) approach. It enables iterative refinement of report sections through PerfReport MCP tools that work together to discover data, save AI-generated revisions, and assemble the final report.

The workflow phases:

1. Discovery — Gather all available data files and determine which sections are revisable
2. AI Generation — Read analysis data and generate revised content for enabled sections
3. Save Revisions — Save each revised section with version tracking
4. Assembly — Assemble the final revised report with AI content replacing placeholders
5. User Review — Present results and gather feedback
6. HITL Iteration — Refine specific sections based on user feedback (repeatable)
7. Confluence Publishing — Optionally publish the revised report

Report Types and Section Differences

This workflow handles two report types with different section IDs:

Single-Run Report Sections:

Comparison Report Sections:

The discover_revision_data tool returns the full list of available sections dynamically. The tables above reflect the known sections defined in report_config.yaml. Additional sections may be added in the future.

Key differences between report types:

• Single-run uses key_observations and issues_table
• Comparison uses key_findings and issues_summary
• Comparison has additional sections: overall_trend_summary and correlation_insights_section
• Single-run has additional sections: jmeter_log_analysis and bottleneck_analysis

Artifact Paths

Single-Run Reports:

artifacts/{run_id}/reports/
├── performance_report_{run_id}.md              # Original report
├── performance_report_{run_id}_original.md     # Backup (after revision)
├── performance_report_{run_id}_revised.md      # AI-revised report
├── report_metadata_{run_id}.json               # Metadata
├── report_metadata_{run_id}_original.json      # Backup metadata
└── revisions/                                  # AI revision files
    ├── AI_EXECUTIVE_SUMMARY_v1.md
    ├── AI_KEY_OBSERVATIONS_v1.md
    └── AI_ISSUES_TABLE_v1.md

Comparison Reports:

artifacts/comparisons/{comparison_id}/
├── comparison_report_*.md              # Original report
├── comparison_report_*_original.md     # Backup (after revision)
├── comparison_report_*_revised.md      # AI-revised report
├── comparison_metadata_*.json          # Metadata
├── comparison_metadata_*_original.json # Backup metadata
├── charts/                             # Comparison charts
└── revisions/                          # AI revision files
    ├── AI_EXECUTIVE_SUMMARY_v1.md
    ├── AI_KEY_FINDINGS_BULLETS_v1.md
    └── AI_ISSUES_SUMMARY_v1.md

Tool Reference

AI Content Guidelines

Executive Summary

• High-level overview of test results
• Include key metrics: success rate, avg response time, throughput
• Highlight critical issues or SLA violations
• 3-5 sentences or bullet points
• Mention test environment and date if relevant
• Incorporate additional_context if provided (project name, purpose)

Key Observations (single_run) / Key Findings (comparison)

• 3-7 key observations or findings
• Bullet points for clarity
• Both positive findings and concerns
• Reference specific APIs or services
• Prioritize by impact
• For comparison: Include trends across runs, scaling behavior, pattern identification

Issues Table (single_run) / Issues Summary (comparison)

• Markdown table: Issue Type, Severity, Count/Impact, Affected Endpoint, Description
• Sort by severity (Critical > High > Medium > Low)
• Include error rates and specific error messages
• For HTTP errors, recommend reviewing JMeter logs
• For comparison: Recurring vs. one-time issues, issues resolved between runs

JMeter Log Analysis (single_run)

• Categorize errors found in JMeter/BlazeMeter logs
• List affected APIs and endpoints
• Correlate with JTL test results data where applicable
• Highlight patterns (e.g., recurring errors, timeouts, connection failures)

Bottleneck Analysis (single_run)

• Identify the concurrency threshold where performance degradation begins
• Severity breakdown by endpoint
• Per-endpoint findings with specific metrics
• Recommendations for addressing bottlenecks

Overall Trend Summary (comparison)

• Narrative describing how key metrics changed across test runs
• Highlight scaling behavior (e.g., linear throughput growth, stable error rates)
• Note any inflection points where performance degraded

Correlation Insights (comparison)

• Performance-infrastructure correlation findings across runs
• CPU/Memory scaling behavior relative to load increases
• Identify which resources correlate most strongly with response time changes

Writing Style

• Professional tone suitable for leadership/stakeholders
• Concise but informative with specific metrics and data points
• Use markdown formatting (headers, bullet points, tables)

Technical Term Definitions

When using terms non-technical stakeholders may not understand:

• 1-2 terms: Add a footnote after Key Observations: > **Terms:** *CoV* — A measure of response time consistency; values >1.0 indicate high variability.
• 3+ terms: Add a "Glossary" section before "Report Generation Details" with a definition table.

Common terms requiring definition: CoV, P90/P95/P99, SLA, Throughput, Latency vs Response Time, Error Rate, mCPU, Connection Pool.

AI Content Guidelines for Comparison Reports

When generating content for comparison reports, consider:

1. Trends Analysis — How performance changed across runs, improvements or degradations, pattern identification
2. Multi-Run Context — User load scaling (e.g., 25 -> 500 VUs), throughput changes, success rate consistency
3. Infrastructure Insights — CPU/Memory scaling behavior, resource utilization patterns, correlation between load and usage
4. Aggregated Issues — Recurring vs. one-time issues, issues resolved between runs, critical patterns

Example comparison executive summary:

"Analysis of 5 test runs (25-750 VUs) shows stable system performance under increasing load. Throughput scaled linearly from 1.76 to 161.54 req/sec while maintaining 99.98%+ success rate. CPU utilization increased proportionally (220 -> 1040 mCPU) indicating healthy resource scaling."

Priority Data Files

For single_run reports (read in this order):

1. artifacts/{run_id}/analysis/performance_analysis.json — Core metrics
2. artifacts/{run_id}/analysis/performance_summary.md — Human-readable summary
3. artifacts/{run_id}/analysis/infrastructure_analysis.json — Infrastructure metrics
4. artifacts/{run_id}/analysis/correlation_analysis.json — Correlations
5. artifacts/{run_id}/analysis/bottleneck_analysis.json — Bottleneck analysis
6. artifacts/{run_id}/analysis/<source>_log_analysis.json — Source: "jmeter" or "blazemeter"
7. artifacts/{run_id}/analysis/log_analysis.json — High-level log analysis incl. Datadog
8. artifacts/{run_id}/blazemeter/test_config.json — Test configuration
9. artifacts/{run_id}/reports/performance_report_{run_id}.md — Current report

For comparison reports (read in this order):

1. artifacts/comparisons/{comparison_id}/comparison_metadata_*.json — Run IDs and config
2. artifacts/comparisons/{comparison_id}/comparison_report_*.md — Current comparison report
3. For each run in run_id_list:
   • artifacts/{run_id}/reports/report_metadata_{run_id}.json — Individual run metrics
   • artifacts/{run_id}/analysis/performance_analysis.json — Detailed metrics per run

Related Rules

• prerequisites.mdc — test_run_id and artifact structure validation
• skill-execution-rules.mdc — Follow steps in order, collect inputs first, do not skip
• mcp-error-handling.mdc — MCP tool error handling (retry policy, reporting format)

Important Notes

• The original report must exist before running the revision workflow.
• Sections must be enabled in report_config.yaml before they can be revised. All sections are disabled by default.
• Each call to prepare_revision_context creates a new version. Previous versions are preserved for comparison.
• The original report and metadata are backed up before any modifications. Backups are not overwritten if they already exist.
• When iterating, provide specific feedback to improve subsequent versions.

Execution

This is an interactive HITL workflow. Steps 6-7 repeat as many times as needed based on user feedback.

Collect Inputs

Ask the user for the following values. Do not proceed until all required values are collected.

REQUIRED:
  run_id      = [test run ID for single_run, or comparison_id for comparison reports]
  report_type = ["single_run" (default) or "comparison"]

OPTIONAL:
  additional_context = [project name, purpose, feature/PBI details from ADO/JIRA]
  sections_to_revise = [user preferences on which sections to revise]

Prerequisites Check

For single_run reports, verify:

• Report exists at artifacts/{run_id}/reports/performance_report_{run_id}.md
• Analysis data exists at artifacts/{run_id}/analysis/
• Sections are enabled in report_config.yaml under revisable_sections.single_run

For comparison reports, verify:

• Report exists at artifacts/comparisons/{comparison_id}/comparison_report_*.md
• Comparison metadata exists at artifacts/comparisons/{comparison_id}/comparison_metadata_*.json
• Individual run reports exist for each run in run_id_list
• Sections are enabled in report_config.yaml under revisable_sections.comparison

Do not proceed until the original report has been generated.

Step 1 — Initialize Task Tracking

Action: Create task items to monitor progress:

• Discovery: Run discover_revision_data
• AI Generation: Read data files and generate content
• Save Revisions: Run prepare_revision_context (per section)
• Assembly: Run revise_performance_test_report
• Review: User reviews revised report
• HITL Iteration (if needed)
• Confluence Publishing (if needed)

Step 2 — Discovery

Input: run_id, report_type, additional_context (optional)

Action: Call MCP tool discover_revision_data

discover_revision_data(
  run_id             = {run_id},
  report_type        = {report_type},
  additional_context = {additional_context}
)

Expected response:

{
  "data_sources": { "blazemeter": [...], "datadog": [...], "analysis": [...], "reports": [...] },
  "revisable_sections": [ { "section_id": "...", "enabled": true/false, ... } ],
  "enabled_section_count": 3,
  "revision_output_path": "artifacts/{run_id}/reports/revisions/",
  "existing_revisions": { ... },
  "revision_guidelines": "..."
}

Save:

• data_sources = file paths organized by MCP source
• revisable_sections = list of sections with enabled/disabled status
• enabled_section_count = number of sections ready for revision

Validation: Verify enabled_section_count > 0. If no sections are enabled, inform the user to update report_config.yaml:

Location: perfreport-mcp/report_config.yaml
Section: revisable_sections.single_run (or .comparison)
Set enabled: true for desired sections

On error: If the tool fails, check run_id, verify the artifacts folder exists, and ensure the original report was generated first. Stop and report to user.

Step 3 — AI Generation

Input: data_sources (from Step 2), report_type, additional_context

Action: Read the priority data files listed in the Reference section to build context.

For each enabled section, generate revised content following the AI Content Guidelines in the Reference section:

Single-run sections:

• executive_summary — 3-5 sentences, key metrics, critical issues
• key_observations — 3-7 bullet points prioritized by impact
• issues_table — Markdown table sorted by severity
• jmeter_log_analysis — Error categorization, affected APIs, JTL correlation findings
• bottleneck_analysis — Degradation thresholds, severity breakdown, per-endpoint findings

Comparison sections:

• executive_summary — Overall comparison findings across runs
• key_findings — Trends and insights, 3-7 bullet points
• issues_summary — Aggregated issues, recurring vs. one-time
• overall_trend_summary — Performance trend narrative across runs
• correlation_insights_section — Performance-infrastructure correlation analysis

Use the data from the analysis files to produce specific, data-driven content.

Save: generated_content = dict mapping section_id to generated markdown content.

Step 4 — Save Revisions

Input: run_id, report_type, generated_content (from Step 3), additional_context

Action: For each enabled section, call MCP tool prepare_revision_context:

prepare_revision_context(
  run_id             = {run_id},
  section_id         = {section_id},
  revised_content    = {content for this section},
  report_type        = {report_type},
  additional_context = {additional_context}
)

Expected response:

{
  "section_full_id": "single_run.executive_summary",
  "revision_number": 1,
  "revision_path": "artifacts/{run_id}/reports/revisions/AI_EXECUTIVE_SUMMARY_v1.md",
  "previous_versions": []
}

Repeat for each enabled section returned by discover_revision_data.

Single-run sections:

1. executive_summary -> save executive summary content
2. key_observations -> save key observations content
3. issues_table -> save issues table content
4. jmeter_log_analysis -> save JMeter log analysis content
5. bottleneck_analysis -> save bottleneck analysis content

Comparison sections:

1. executive_summary -> save executive summary content
2. key_findings -> save key findings content
3. issues_summary -> save issues summary content
4. overall_trend_summary -> save overall trend summary content
5. correlation_insights_section -> save correlation insights content

Only process sections where enabled: true in the discovery response.

Save: Track which sections were saved and their version numbers.

On error: If the tool fails, check section_id is valid, content is not empty, and report_type matches the run_id. Stop and report to user.

Step 5 — Assembly

Input: run_id, report_type

Action: Call MCP tool revise_performance_test_report

revise_performance_test_report(
  run_id           = {run_id},
  report_type      = {report_type},
  revision_version = {optional — defaults to latest}
)

The tool:

1. Backs up original report to *_original.md
2. Backs up metadata to *_original.json
3. Replaces placeholders with AI-revised content
4. Saves revised report to *_revised.md
5. Updates metadata with revision info

Expected response:

{
  "status": "success",
  "revised_report_path": "artifacts/{run_id}/reports/performance_report_{run_id}_revised.md",
  "backup_report_path": "artifacts/{run_id}/reports/performance_report_{run_id}_original.md",
  "sections_revised": ["executive_summary", "key_observations", "issues_table"],
  "revision_versions_used": { "executive_summary": 1, "key_observations": 1, "issues_table": 1 },
  "warnings": []
}

Save: revised_report_path, sections_revised, revision_versions_used.

Validation: Verify status is "success", sections_revised contains all expected sections, revised_report_path exists. Review any warnings.

On error: Verify revision files exist in the revisions folder and at least one section is enabled. Stop and report to user.

Step 6 — User Review

Input: revised_report_path, sections_revised, revision_versions_used

Action: Present the results to the user:

1. Display the path to the revised report
2. Summarize which sections were revised
3. Show the version numbers used
4. Note any warnings from the assembly

Ask the user:

• "Would you like to review the revised report?"
• "Are there any sections that need further refinement?"
• "Should I make any adjustments to the content?"

If the user is satisfied, go to Step 8. If the user wants changes, go to Step 7.

Step 7 — HITL Iteration (repeatable)

When: The user requests changes to specific sections.

7a. Gather Feedback

Collect from the user:

• Which section needs revision?
• What changes are needed?
• Any additional context to incorporate?

7b. Re-generate Content

Generate new content for the specified section incorporating the user's feedback.

7c. Save New Version

prepare_revision_context(
  run_id             = {run_id},
  section_id         = {section_id},
  revised_content    = {new content},
  report_type        = {report_type},
  additional_context = {feedback context}
)

The tool automatically increments the version number (e.g., AI_EXECUTIVE_SUMMARY_v2.md).

7d. Re-assemble Report

revise_performance_test_report(
  run_id           = {run_id},
  report_type      = {report_type},
  revision_version = {optional — specify version or omit for latest}
)

7e. Present Updated Results

Go back to Step 6 to present the updated report and ask for feedback.

Continue the HITL loop until the user approves the revised report.

Step 8 — Final Output Summary

Action: Present a final summary to the user.

File Locations

For single_run reports:

• Original (backed up): artifacts/{run_id}/reports/performance_report_{run_id}_original.md
• Revised: artifacts/{run_id}/reports/performance_report_{run_id}_revised.md
• Revision files: artifacts/{run_id}/reports/revisions/
• Metadata: artifacts/{run_id}/reports/report_metadata_{run_id}.json

For comparison reports:

• Original (backed up): artifacts/comparisons/{comparison_id}/comparison_report_*_original.md
• Revised: artifacts/comparisons/{comparison_id}/comparison_report_*_revised.md
• Revision files: artifacts/comparisons/{comparison_id}/revisions/
• Metadata: artifacts/comparisons/{comparison_id}/comparison_metadata_*.json

Revision Summary Table (example for single_run)

Next Steps

Ask the user:

• "Would you like to publish the revised report to Confluence?" — If yes, go to Step 9.
• "Would you like to generate additional charts?" — Use create_chart from PerfReport MCP.
• "Would you like to create a comparison report?" — Follow the skill at .cursor/skills/comparison-report-workflow/SKILL.md.

Step 9 — Publish Revised Report to Confluence (Optional)

When: The user wants to publish the revised report to an existing Confluence page.

Prerequisites:

• Revised report exists (*_revised.md)
• page_ref (page ID) from the initial publish is available
• Images have already been attached during the initial publish (no re-attach needed)
• confluence_mode = "cloud" or "onprem"

9a. Convert Revised Markdown to XHTML

For single_run reports:

convert_markdown_to_xhtml(
  test_run_id = {run_id},
  filename    = "performance_report_{run_id}_revised.md",
  report_type = "single_run"
)

For comparison reports:

convert_markdown_to_xhtml(
  test_run_id = {comparison_id},
  filename    = "comparison_report_{ids}_revised.md",
  report_type = "comparison"
)

This creates a *_revised.xhtml file in the reports folder.

9b. Update Confluence Page with Revised Content

For single_run reports:

update_page(
  page_ref    = {page_ref},
  test_run_id = {run_id},
  mode        = {confluence_mode},
  use_revised = true,
  report_type = "single_run"
)

For comparison reports:

update_page(
  page_ref    = {page_ref},
  test_run_id = {comparison_id},
  mode        = {confluence_mode},
  use_revised = true,
  report_type = "comparison"
)

The update_page tool with use_revised=true:

1. Selects the *_revised.xhtml file (not the main XHTML)
2. Substitutes chart placeholders with embedded <ac:image> markup
3. Creates a *_revised_with_images.xhtml file for reference
4. Updates the Confluence page with the revised content

Images are already attached from the initial publish. Only placeholder substitution and page content update are needed.

9c. Verification

Verify:

• Response shows "used_revised": true
• xhtml_source field shows the *_revised.xhtml filename
• Chart placeholders were replaced (check placeholders_replaced)
• Page URL is accessible and shows the revised content

On error: If API calls fail, retry up to 3 times. Report error to user.

Error Handling

These rules apply to every step:

• PerfReport MCP tools (discover, prepare, revise): These are Python code executions. Do NOT retry on failure. Report the error with full details. Do NOT attempt to fix code.
• Confluence MCP tools (convert, update_page): These are API calls. Retry up to 3 times. Wait 5-10 seconds between retries.
• If any step fails, stop and report to the user with:
  • Which tool failed
  • Full error message
  • What was expected vs. what happened
• Do NOT proceed to the next step if the current step failed.
• Ask the user for next steps on any error.