execdiff 0.0.4

Passive execution tracing for file and package changes.

ExecDiff

See what AI-generated code will change before running it.

---

Problem

AI coding tools and agents today can:

• install dependencies
• create files
• modify configs
• run migrations
• delete project files

All automatically.

When something breaks after execution, tools cannot answer:

What exactly changed because of this action?

Git tracks source code changes —
but it does not track execution side effects like:

• newly installed Python packages
• runtime-created files
• deleted files
• modified configs

So tools often fall back to:

regenerate and try again

---

Solution

ExecDiff allows tools to run AI-generated code and observe:

what changed in the workspace because of that execution

It detects:

• files created
• files modified
• files deleted
• Python packages installed

inside a specific workspace
during a specific execution window.

---

Installation

pip install execdiff

---

Example

Create a test script:

import execdiff
import json
import os

os.makedirs("workspace", exist_ok=True)

diff = execdiff.run_traced(
    "touch workspace/test.txt",
    workspace="workspace"
)

print(json.dumps(diff, indent=2))

Run:

python test.py

---

API Reference

start_action_trace(workspace=".")

Start tracing a workspace for changes. Must be called before any operations.

import execdiff

execdiff.start_action_trace(workspace="./my_workspace")
# ... your code that makes changes ...

stop_action_trace()

Stop tracing and return a diff of all changes detected. Automatically logs to .execdiff/logs/actions.jsonl.

diff = execdiff.stop_action_trace()
# Returns: {"files": {...}, "packages": {...}}

last_action_summary(workspace=".")

Get a human-readable summary of the last action trace without parsing JSON.

summary = execdiff.last_action_summary(workspace=".")
print(summary)

Output example:

Last AI Action:

Created:
- output.txt
- data.json

Installed:
- requests==2.32.0

snapshot_workspace_state(workspace)

Take a full metadata snapshot of the workspace (files with mtime/size, installed packages).

state = execdiff.snapshot_workspace_state(workspace=".")
# Returns: {"files": {...}, "packages": {...}}

---

Output Format

Diff Structure

{
  "files": {
    "created": [{"path": "file.txt", "mtime": 123.45, "size": 1024}],
    "modified": [{"path": "config.yaml", "before_mtime": 123, "after_mtime": 124, "before_size": 512, "after_size": 1024}],
    "deleted": [{"path": "old_file.txt", "mtime": 123.45, "size": 256}]
  },
  "packages": {
    "installed": [{"name": "requests", "version": "2.32.0"}],
    "upgraded": [{"name": "django", "before_version": "3.2", "after_version": "4.0"}],
    "removed": [{"name": "deprecated_lib", "version": "1.0"}]
  }
}

Log File

All action traces are automatically persisted to .execdiff/logs/actions.jsonl:

{
  "timestamp": "2026-02-18T18:19:35.872838",
  "workspace": "/path/to/workspace",
  "diff": {...}
}

---

Use Cases

ExecDiff can help AI coding tools:

• preview changes before applying generated code
• detect unintended file or dependency changes
• explain execution impact to users
• debug failed automation
• build undo / rollback systems

---

License

MIT