Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use Proxyman MCP to inspect and debug captured HTTP/HTTPS/WebSocket traffic and operate Proxyman through all MCP actions. Use this skill when an agent needs to analyze traffic, diagnose missing capture, manage rules, use Compose, inspect WebSockets, install certificates, control proxy settings, use setup guidance, export flows, generate code, run Proxyman automations, or troubleshoot Proxyman MCP.
.claude/skills/proxymanapp-proxyman-traffic-debugging/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-15 | ✗→✓ | ▲ Improved | 112% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 82% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 224% | 0% |
| case-12 | ✓→✓ | = Same ✓ | 157% | 0% |
| case-17 | ✓→✓ | = Same ✓ | 201% | 0% |
Use Proxyman MCP as a local control plane for the running Proxyman app. Start read-only, identify the relevant captured flow or current state, then make only the change the user requested.
If Proxyman tools are unavailable, use proxyman-mcp-setup. If Proxyman is absent, use proxyman-download-setup. For target-specific HTTPS configuration, use proxyman-https-capture. For current product documentation, use proxyman-debugging-tools. For shell automation, use proxyman-cli.
Use this order whenever sources disagree:
tools/list, resources/list, and prompts/list for available MCP capabilities and exact schemas.proxyman-debugging-tools for GUI behavior and features beyond the built-in setup index.Never invent a tool, enum, parameter, rule ID, flow ID, file path, or command flag. The stdio bridge uses strict schemas.
mcp-server over stdio. The bridge reads a local handshake and forwards calls to an authenticated localhost server in the running app..app injection, and app control are primarily macOS operations. Trust live discovery on Windows and Linux.get_version to verify the bridge reaches Proxyman.get_proxy_status to learn recording state, proxy port, and SSL Proxying state.get_flows, filter_flows, a status call, or the relevant list call.For tool selection, read MCP tool catalog. For multi-step tasks, read MCP workflows. For Breakpoint, Map Local, Map Remote, Scripting, or overlapping modification rules, read request and response rules. For built-in knowledge, read MCP resources and prompts.
Keep Settings > MCP > Redact Sensitive Data Before Sending to AI enabled by default. It covers common secret headers, cookies, query values, JWTs, credentials, and secret-like body fields in MCP output.
Redaction does not sanitize the original HAR or Proxyman log written by export_flows. Treat exported files as sensitive.
Require explicit user approval immediately before:
Use an explicit approved destination for exports. Do not claim an export is redacted merely because MCP previews are redacted.
get_proxy_status.get_flows or filter_flows. Use client/device identity when the user names a source.get_flow_detail; use its Summary metadata, timing, sizes, connection/TLS details, headers, query, cookies, and body preview.tool not found result.proxyman-cli and say that MCP cannot perform it in this build.get_flows/filter_flows return at most 500 results and body previews can be truncated. Use a user-approved export for complete archival data.filter_flows can filter on client; prefer it when distinguishing device, simulator, Atlantis, or app sources.list_rules for Breakpoint, Map Local, Map Remote, Block List, Scripting, Allow List, Network Condition, and DNS Spoofing. Reverse Proxy has its own list tool.QUERY; use ANY only for all methods. Set include_paths deliberately because defaults differ by rule type.list_tool_status.response_body, status_code, and content_type together.send_compose_http is HTTP-only; WebSocket URLs are rejected.close_websocket_session applies only to an active Compose-created WebSocket session.docs://scripting/snippet-code before non-trivial script generation.get_proxyman_cli_help only returns safe, version-matched help. It never executes an operational CLI command.Finish with:
Do not claim traffic is complete when capture state, time window, client source, or preview truncation makes the result inconclusive.
proxyman-mcp-setup.proxyman-https-capture.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 15,126 | 11,431 | -24% | 1 | 1 | 0% | 2,161 | 2,431 | +12% | 0 | 0 | — |
case-02 | fail→fail | 6,679 | 4,382 | -34% | 1 | 1 | 0% | 567 | 2,434 | +329% | 0 | 0 | — |
case-03 | fail→fail | 6,852 | 6,544 | -4% | 1 | 1 | 0% | 248 | 2,573 | +938% | 0 | 0 | — |
case-04 | fail→fail | 10,068 | 3,438 | -66% | 1 | 1 | 0% | 1,508 | 2,418 | +60% | 0 | 0 | — |
case-05 | fail→fail | 9,973 | 3,347 | -66% | 1 | 1 | 0% | 1,459 | 2,344 | +61% | 0 | 0 | — |
case-06 | fail→fail | 10,317 | 5,922 | -43% | 1 | 1 | 0% | 1,502 | 2,648 | +76% | 0 | 0 | — |
case-07 | fail→fail | 9,671 | 4,267 | -56% | 1 | 1 | 0% | 1,693 | 2,413 | +43% | 0 | 0 | — |
case-08 | fail→fail | 8,921 | 5,085 | -43% | 1 | 1 | 0% | 1,389 | 2,475 | +78% | 0 | 0 | — |
case-09 | fail→fail | 14,124 | 6,400 | -55% | 1 | 1 | 0% | 2,241 | 2,398 | +7% | 0 | 0 | — |
case-10 | fail→fail | 10,448 | 4,745 | -55% | 1 | 1 | 0% | 1,640 | 2,488 | +52% | 0 | 0 | — |
case-11 | fail→fail | 9,105 | 4,279 | -53% | 1 | 1 | 0% | 1,622 | 2,522 | +55% | 0 | 0 | — |
case-12 | pass→pass | 11,971 | 21,465 | +79% | 1 | 1 | 0% | 1,781 | 4,586 | +157% | 0 | 0 | — |
case-13 | fail→fail | 10,766 | 4,333 | -60% | 1 | 1 | 0% | 1,625 | 2,468 | +52% | 0 | 0 | — |
case-14 | fail→fail | 4,269 | 5,215 | +22% | 1 | 1 | 0% | 650 | 2,540 | +291% | 0 | 0 | — |
case-19 | fail→fail | 7,991 | 5,034 | -37% | 1 | 1 | 0% | 1,247 | 2,584 | +107% | 0 | 0 | — |
case-15 | fail→pass | 10,371 | 7,830 | -25% | 1 | 1 | 0% | 1,559 | 3,298 | +112% | 0 | 0 | — |
case-16 | fail→pass | 14,353 | 14,737 | +3% | 1 | 1 | 0% | 2,193 | 3,995 | +82% | 0 | 0 | — |
case-17 | pass→pass | 8,920 | 13,435 | +51% | 1 | 1 | 0% | 1,419 | 4,271 | +201% | 0 | 0 | — |
case-18 | fail→pass | 6,218 | 6,695 | +8% | 1 | 1 | 0% | 980 | 3,180 | +224% | 0 | 0 | — |
case-20 | pass→pass | 9,635 | 7,684 | -20% | 1 | 1 | 0% | 1,626 | 3,436 | +111% | 0 | 0 | — |
case-21 | fail→fail | 13,309 | 16,570 | +25% | 1 | 1 | 0% | 2,450 | 4,491 | +83% | 0 | 0 | — |
case-22 | fail→fail | 7,647 | 5,628 | -26% | 1 | 1 | 0% | 1,219 | 2,992 | +145% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 8 counted toward the lift figure. The other 14 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +14 percentage points is the difference between those two pass rates over the 8 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/3/2026 | +50% |
| gemini-3.6-flash | verified | 8/3/2026 | +32% |
Other measured skills in the registry, with their headline benchmark lift.