# Edt MCP Project Profiling

> Run a bounded 1C performance profiling experiment through EDT-MCP and collect attributable evidence without leaving profiling active.

- Skill: `ditrixnew/edt-mcp-project-profiling` (Agent Skill)
- Install (CLI): `npx skillmds@latest add ditrixnew/edt-mcp-project-profiling`
- Raw SKILL.md: https://api.skillmd.com/api/skills/ditrixnew/edt-mcp-project-profiling/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: ditrixnew (https://skillmd.com/u/ditrixnew)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/ditrixnew/edt-mcp-project-profiling

---


# EDT-MCP profiling

## Purpose and trigger

Use this skill when one reproducible performance question requires runtime
measurement against an exact 1C application.

## Operating rule

Read and apply [the common operating rules](../COMMON.md) before this workflow.

## Task boundary

Measure one authorized scenario and data volume. Do not mix the measurement
with refactoring, broad load generation, or unrelated runtime activity.

## Primary workflow

1. Define the question and resolve an active, unambiguous debug target with
   `debug_status`; route launch or Attach preparation to
   `edt-mcp-project-runtime-debug` when needed.
2. Establish an attributable profiling window using current help and existing
   `get_profiling_results` state. Keep the global profiling surface quiescent
   through the final result read; if result identity cannot be matched
   independently, treat the global result as unattributed.
3. Call `start_profiling` and claim ownership only when its result confirms
   this task started the profiling window. If profiling was already active, do
   not stop or replace it; stop and report that an attributable window could
   not be established without disrupting another owner. Otherwise execute only
   the bounded scenario and call `stop_profiling` for that task-owned window on
   success, failure, timeout, or interruption.
4. Read `get_profiling_results` in that protected window, correlate candidate
   methods/lines with exact source, and repeat only for a controlled comparison.
5. Treat returned profiling rows as potentially partial; do not make absolute
   hotspot or completeness claims unless the measurement proves them.

## Authority rule

The target application, scenario, data effects, launch/Attach work, and any
repeat measurement must be authorized. Profiling does not authorize a code fix.

## Stop rule

Stop and clean up when the target changes, another session prevents reliable
attribution, the scenario would mutate prohibited data, or results cannot be
linked to the intended run.

## Completion signal

Report the exact target and scenario, conditions, observed duration and
candidate hotspots, partial-result and attribution limits, cleanup status, and
the next validation experiment for any proposed optimization.