HAR to JMeter JMX Conversion

When to Use This Skill

• User wants to convert a HAR file into a JMeter script
• User mentions HAR, Chrome DevTools, Fiddler, Charles, mitmproxy, or Postman recording
• User has a .har file from a browser recording and needs a .jmx load test script

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

1. Converts a HAR file into a structured network capture JSON
2. Analyzes the capture to identify dynamic values (correlations) and auto-generates variable names for them
3. (Gate) If low-confidence or orphan correlations are found, pauses for correlation variable name review before script generation
4. Generates a parameterized JMeter JMX script from the finalized naming data

Parameter Reference

Required:

• test_run_id — Unique identifier for this test run (e.g., hardemo1). All artifacts are written to artifacts/{test_run_id}/jmeter/.
• har_path — Absolute filesystem path to the .har file. Relative paths will fail.

Optional:

• step_strategy — How to group HAR entries into test steps. Defaults to auto.
  • auto — Use HAR page references when available, fall back to time-gap detection
  • page — Group strictly by HAR page entries
  • time_gap — Group by time gaps between requests
  • single_step — All entries in one step
• time_gap_threshold_ms — Gap threshold in milliseconds for the time_gap strategy. Defaults to 3000.

Related Rules

These Cursor Rules apply when using this skill:

• 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 — Error handling for MCP tool calls (retry policy, reporting format)
• jmeter-script-guardrails.mdc — Applies to downstream HITL editing and debugging after conversion

Notes

• HAR files from real browser sessions contain actual request/response data, so correlation analysis typically finds more meaningful results than synthetic Swagger-based captures.
• Use page strategy if the HAR has clean page entries. Use time_gap if pages are missing.
• Cache-busting parameters (e.g., _t, _) with epoch-millisecond values are common. These are typically orphan IDs best handled with JMeter's __time() function.
• OAuth redirect flows may have tokens exchanged outside the capture window, resulting in orphan correlations for auth-related values.
• After conversion, the user can proceed to HITL editing or debugging workflows.
• The correlation naming file is auto-generated by the analysis tool. Always review and adjust variable names before generating the JMX script when the analysis reports any low-confidence or orphan correlations — variable names are baked into the JMX at generation time and cannot be updated retroactively without re-generating or manually editing the script. Use the correlation naming skill at .cursor/skills/jmeter-correlation-naming/SKILL.md (Scenario A) for the review.

Execution

Follow these steps exactly, in order. Each step has one action.

Collect Inputs

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

REQUIRED:
  test_run_id = [ask user]
  har_path    = [ask user — must be absolute path]

OPTIONAL:
  step_strategy        = [ask user or use default: "auto"]
  time_gap_threshold_ms = [ask user or use default: 3000]

Step 1 — Convert HAR to Network Capture

Input: test_run_id, har_path, step_strategy, time_gap_threshold_ms

Action: Call MCP tool convert_har_to_capture

convert_har_to_capture(
  test_run_id           = {test_run_id},
  har_path              = {har_path},
  step_strategy         = {step_strategy},
  time_gap_threshold_ms = {time_gap_threshold_ms}
)

Expected response:

{
  "status": "OK",
  "network_capture_path": "artifacts/{test_run_id}/jmeter/network-capture/network_capture_YYYYMMDD_HHMMSS.json",
  "test_run_id": "{test_run_id}"
}

Save: capture_path = value of network_capture_path from the response.

On error: If status is "ERROR", stop. Report the full error message to the user.

Step 2 — Analyze Network Traffic

Input: test_run_id

Action: Call MCP tool analyze_network_traffic

analyze_network_traffic(
  test_run_id = {test_run_id}
)

Expected response:

{
  "status": "OK",
  "correlation_spec_path": "artifacts/{test_run_id}/jmeter/correlation_spec.json",
  "correlation_naming_path": "artifacts/{test_run_id}/jmeter/correlation_naming.json",
  "count": 7,
  "summary": { "total_correlations": 7, "business_ids": 3, "..." : "..." }
}

Save: correlation_count = value of count from the response.

On error: If status is "ERROR", stop. Report the full error message to the user.

This step produces two files:

• correlation_spec.json — raw correlation analysis
• correlation_naming.json — auto-generated variable names

Step 2a — Correlation Naming Review (Conditional Gate)

Condition: Check the summary from Step 2. If any of the following are non-zero, this gate is mandatory before proceeding to Step 3:

• low_confidence > 0
• orphan_ids > 0

Action: Pause and present the low-confidence and orphan correlations to the user.

For each flagged correlation, show:

• The auto-generated variable_name
• The type and value_type (e.g. orphan_id / email, progressive_auth / jwt)
• Where the value was found (source URL and location)
• A suggested rename if the auto-generated name is generic or ambiguous

Ask the user to confirm or provide corrected names before continuing.

Once names are confirmed, follow the correlation naming skill at .cursor/skills/jmeter-correlation-naming/SKILL.md (Scenario A) to apply any changes to correlation_naming.json.

If all correlations are high or medium confidence with no orphans: skip this gate and proceed directly to Step 3.

Why this gate exists: generate_jmeter_script reads correlation_naming.json at generation time. Variable names are baked into extractor names and parameterized request fields in the JMX. Renaming after generation only updates the naming file — it does not retroactively update the script.

Step 3 — Generate JMeter JMX Script

Input: test_run_id, capture_path (saved from Step 1)

Action: Call MCP tool generate_jmeter_script

generate_jmeter_script(
  test_run_id = {test_run_id},
  json_path   = {capture_path}
)

Expected response:

{
  "status": "success",
  "jmx_path": "artifacts/{test_run_id}/jmeter/ai-generated_script_YYYYMMDD_HHMMSS.jmx",
  "message": "JMX script generated successfully: ..."
}

Save: jmx_path = value of jmx_path from the response.

On error: If status is "error", stop. Report the full error message to the user.

Note: This tool returns lowercase "success" / "error", unlike the other tools which return uppercase "OK" / "ERROR".

Step 4 — Report to User

Input: test_run_id, jmx_path (saved from Step 3), correlation_count (saved from Step 2)

Action: Present the results to the user.

Tell the user:

• The JMX script was created at {jmx_path}
• {correlation_count} correlations were detected and parameterized
• The following artifacts were generated:

artifacts/{test_run_id}/jmeter/
├── network-capture/
│   └── network_capture_<timestamp>.json
├── capture_manifest.json
├── correlation_spec.json
├── correlation_naming.json
├── ai-generated_script_<timestamp>.jmx
└── testdata_csv/
    └── environment.csv

Ask the user:

• "Do you want to proceed with HITL editing or debugging?" — If yes, proceed to the appropriate downstream workflow.

Error Handling

These rules apply to every step:

• If any MCP tool returns an error status, stop immediately.
• Report the full error message to the user.
• Do NOT write code to fix MCP tool issues.
• Do NOT proceed to the next step if the current step failed.
• Ask the user for next steps.