# Edt MCP Project Code Research

> Investigate 1C BSL behavior, dependencies, impact, and implementation differences through EDT-MCP. Read-only by default; not for plugin development.

- Skill: `ditrixnew/edt-mcp-project-code-research` (Agent Skill)
- Install (CLI): `npx skillmds@latest add ditrixnew/edt-mcp-project-code-research`
- Raw SKILL.md: https://api.skillmd.com/api/skills/ditrixnew/edt-mcp-project-code-research/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-code-research

---


# EDT-MCP code research

## Purpose and trigger

Use this skill to produce a bounded, evidence-backed explanation of 1C project
code, dependencies, impact, or implementation differences.

## Operating rule

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

## Task boundary

Keep the task read-only and tied to exact project, metadata FQN, module, method,
or symbol targets. Route a known bounded correction to
`edt-mcp-project-local-fix`; do not turn research into implementation without
authorization.

## Primary workflow

1. Resolve candidates with `get_metadata_objects`, `list_modules`, and
   `search_in_code`. Search is literal and not dialect-aware: for identifiers,
   use AST-backed definition/reference/hierarchy tools and do not infer absence
   from one Russian or English spelling.
2. Narrow with `get_module_structure`, then prefer `read_method_source` over
   `read_module_source` unless module-level context is required.
3. Follow only the relationships needed using `go_to_definition`,
   `find_references`, `get_method_call_hierarchy`, `get_outgoing_structures`,
   or `get_symbol_info`.
4. `get_method_call_hierarchy` scans only one project's `<project>/src`.
   Inspect each relevant base/extension project separately and name any project
   left unsearched. Treat `get_outgoing_structures` as a heuristic lower bound,
   even when its response is not marked partial.
5. Treat every single-hop/depth-1 result and structured-output analysis as a
   lower bound. Inspect bounded source and relevant projects before claiming a
   complete contract.
6. Start with counts, file lists, filters, or method/range reads; request full
   payloads or more pages only when needed. Preserve truncation and cursor
   evidence in the result.
7. Use `get_platform_documentation` when the conclusion depends on platform
   behavior rather than project code.

## Authority rule

Do not write code, metadata, runtime data, or Git state. Request a new task
boundary before any implementation or runtime experiment.

## Stop rule

Stop when the exact target cannot be resolved, required project evidence is
unavailable, or the conclusion needs unauthorized runtime proof.

## Completion signal

Return exact targets, the evidenced call/data flow, relevant dependencies and
exceptions, impact classification, partial-result caveats, unknowns, and the
smallest useful next validation step.