# Omp Ida Native

> Analyze executables and IDA databases in OMP using its built-in ida actions, binary read views, and persistent Python helpers. Use for native decompilation, evidence snapshots, candidate function triage, version comparison, dynamic observation plans, or database edits when the native IDA interface is available.

- Skill: `2233admin/omp-ida-native` (Agent Skill, multi-file: 10 files)
- Install (CLI): `npx get-skill add 2233admin/omp-ida-native`
- Raw SKILL.md: https://api.skillmd.com/api/skills/2233admin/omp-ida-native/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: 2233admin (https://skillmd.com/u/2233admin)
- Updated: 2026-11-11
- Page: https://skillmd.com/skills/2233admin/omp-ida-native

---


# OMP native IDA analysis

## ACTION REQUIRED

1. `NOW`: Identify the input, requested result, permitted database edits, and a check that proves the result.
2. `NOW`: Inspect the available IDA interface. The native schema has `action` values `list`, `open`, `save`, `close`, `exec`, `rename`, `comment`, `set_type`, `make_function`, with `db` and `code` parameters. If OMP exposes it through `xd://ida`, read that interface before invoking it.
3. `NEXT`: Call native `ida` with `{"action":"list"}`. Reuse the matching database or open the input; retain its returned native database ID.
4. `ACT`: Read the overview, select relevant functions, then perform the smallest query or batch that answers the task.

If the native interface is absent or a dependency probe fails, read [setup](references/setup.md). An installed `omp` executable or IDA directory alone does not prove this interface works. When an official Hex-Rays extension exposes `ida_*` MCP tools, use that extension's live schema and guidance; this native workflow does not apply.

## Workflow

### Open and inspect

Call `ida` with `{"action":"open","db":"sample.exe"}` or read `sample.exe` for an overview. Executables use a managed database; existing `.i64`/`.idb` files open in place and may be saved there.

Use `read` views for observations: `sample.exe:main:1-80` for pseudocode, `sample.exe:main:asm` for assembly, `sample.exe:strings`, or `sample.exe:xrefs:main`. Read [native interface examples](references/native-ida.md) when selecting a view, batch helper, or edit operation.

Always pass the returned native `db` ID to subsequent `ida` calls, including when only one database is open. Each database has its own Python namespace, shared with other agents in this project. Coordinate edits and use task-specific variable names.

### Query in bounded batches

Prefer a relevant regex and a small result slice. Combine related queries in one `exec` call when they address the same task. The helpers return Python values; print compact JSON when collecting machine-readable evidence. Record what was filtered or truncated.

Persistent variables are a convenience during analysis. Save durable evidence in the task report: input identity, native database ID, addresses, actual outputs, and remaining unknowns. Reopening a closed database resets the namespace.

### Reusable analysis workflows

| Requested result | Entry | Output |
| --- | --- | --- |
| Candidate functions or frozen observations | [Snapshot and triage](references/triage.md) | Source SHA, EA/RVA, bounded evidence and coverage |
| Two-version observations | [Version comparison](references/version-diff.md) | Explicit/unique-name correspondences, changed fields and unmatched items |
| Runtime verification plan | [Dynamic observation plan](references/dynamic-validation.md) | Bounded Frida script with unresolved identity/trigger preconditions |

These helpers have scoped checks on self-authored x64 PE fixtures: real native snapshot/triage, captured-version comparison and generated Frida-script observations. See each workflow's evidence link for limits; other inputs/architectures remain unverified. Preserve complete JSON artifacts before passing them between workflows. Scripts stay within this portable directory. Generating a dynamic plan does not execute it or establish that its hypothesis is true.

### Edit and save

Use native `rename`, `comment`, `set_type`, or `make_function` for permitted database edits. Generated `read` views are immutable text; edit the database through `ida`.

After edits, explicitly `save`, re-read the affected view, and compare it with the requested change. Automatic saving is a recovery aid. Close a database only when this task owns its lifecycle and other users have finished with it; `close` saves by default.

### Handle resource and failure limits

OMP shares database daemons across project agents. Defaults are four open databases, automatic saving after about ten idle seconds when dirty, and saving/closing after fifteen idle minutes. Opening beyond the limit evicts an idle database; all-busy requests fail. Live settings and tool descriptions take precedence over these versioned defaults.

Keep `exec` finite; its default timeout is 120 seconds and applies to the queue wait as well. A queued timeout does not cancel another caller's running request. Report failures and retry only after diagnosing them. Worker termination can lose edits since the last save; verify saved state before continuing.

## 任务完成自检

Before claiming completion, MUST confirm:

- The result comes from the requested input and its actual native database ID.
- Function/address evidence satisfies the task's check; fallback assembly is identified as assembly.
- Permitted edits were saved and checked; no other task's database was closed or discarded.
- The report names omitted results, failures, unresolved questions, and checks not run.

This skill uses OMP's existing IDA implementation. Backend contracts and setup are in the linked references; it does not install or register a separate MCP service.

