AGENTS.md - basic FFAStrans knowledge for Agents

[emcodem]

For the users of claude code and similar tools:
I do keep this collection of generic ffastrans system knowledge in the ffastrans install dir so AGENTS.md and docs folder is next to ffastrans.exe.

The goal of providing such knowledge files is that Agents don't need to re-collect all basic knowledge from scratch each time you start a new session (or make you independent of the Agents own memory). Ultimately, keeping short generic .md files like this to guide Agents shall help to limit Token and Context usage and therefore speed up and cause less cost.
For example: if you ask the agent what went wrong with this %job_id%, it might by default have to check lots of files and folders just to find out where the job logs are and how to parse them efficiently. All of this can be saved when the agent reads the generic knowledge file up in front.

For ffastrans 1.4:

AGENTS.zip

(28.43 KiB) Downloaded 3 times

(updated 13.06, removed misleading o_media)

[Julian]

This is awesome, thank you!

[emcodem]

Some more infos on how to efficiently use this.

Up in front: the following is about Commandline AI tools that can access your local files, the most prominent mentioned below. In contrast: The Browser based assistant tools work differently but they are pretty useless because Browser cannot access Local files. For Supporting FFAStrans the Agents must be able to read config and log files.
The problem is: each AI tool loads this documentation differently. Here are some tipps from my side:
Launch the assistant from Your FFAStrans installatoin Folder
Launching your AI assistant in the correct folder comes with benefits. For example:

• File Access for Files in subfolders does not always need manual OK

• Some Tools automatically read special .md Files in the project root which can lead to a quickstart

These tools look for their documentation file in the folder where you launch them, not where the tool is installed. If you open a terminal and run the tool from your home directory or Downloads folder, it won't find AGENTS.md — and it won't warn you either. It just starts without the FFastrans context, silently.
Navigate into FFAStrans install folder before starting the Assistant:

Code: Select all

cd C:\FFAStrans
or if on UNC:
pushd \\server\share\FFAStrans

Of course you should have a batch on the desktop that changes directory and starts your tool automatically in the correct folder e.g. FFASTRANS_Support.bat.
Wrong — launching from anywhere else:

Code: Select all

C:\Users\You>claude

How to tell if it worked: At the start of the session, ask the AI: "What do you know about FFastrans?" If it gives a blank or generic answer, it hasn't loaded the docs. Double-check you're in the right folder and the file is there.
Want to skip this entirely? Use the global config option where available (Claude Code and Gemini CLI both support it — see their sections below). This loads the docs for every session on your machine, regardless of which folder you're in.

Tool-by-Tool Setup

Claude Code
Claude Code looks for a file called CLAUDE.md in your project folder. Just rename (or copy) our AGENTS.md to CLAUDE.md and place it in your working directory.

Code: Select all

FFAStrans/

├── CLAUDE.md          ← rename AGENTS.md to this

└── docs/              ← keep the docs subfolder here too

Claude will read it automatically every time you start a session in that folder. You can also put a CLAUDE.md in your home folder (%USERPROFILE%.claude\CLAUDE.md) if you want it loaded globally for all projects.

Codex (OpenAI)
Codex uses AGENTS.md — which means our file works as-is, no renaming needed.

Code: Select all

FFAStrans/

├── AGENTS.md          ← already the right name

└── docs/              ← keep the docs subfolder here too

Drop it in FFAStrans root and Codex will pick it up automatically at the start of each session.

Gemini CLI
Gemini CLI looks for a file called GEMINI.md. Copy our AGENTS.md and rename it.

Code: Select all

FFAStrans/

├── GEMINI.md          ← rename AGENTS.md to this

└── docs/              ← keep the docs subfolder here too

For a global version that loads for every project on your machine, place it at %USERPROFILE%.gemini\GEMINI.md.

GitHub Copilot CLI
The CLI version of Copilot doesn't auto-load a file the way the others do. You have two options:
Option A — Paste it in at the start of your session:

Just open AGENTS.md, copy the contents, and paste it as your first message to Copilot before asking anything. Something like: "Here's documentation for the tool I'm working with: [paste]"
Option B — Use a batch file to inject it automatically:

Save the following as copilot-ffastrans.bat in your FFAStrans folder or anywhere in your PATH:

Code: Select all

@echo off

set /p PROMPT="What do you need help with? "

gh copilot suggest "%PROMPT% --- Context: $(type %USERPROFILE%\ffastrans\AGENTS.md)"

Bonus: If you use Copilot inside VS Code or another IDE (not the CLI), you can place a file at .github/copilot-instructions.md in FFAStrans and it will be loaded automatically in the editor.

Kimi Code (Moonshot)
Kimi Code doesn't have an auto-load file standard yet, but you can pass documentation directly using the --context flag:

Code: Select all

kimi --context /path/to/AGENTS.md --prompt "help me set up a transcode workflow"

If you're using Kimi interactively, paste the AGENTS.md content at the start of your session before asking questions — same approach as Copilot CLI Option A above.
OpenCode
OpenCode is configured via a file called opencode.json in FFAStrans root. You can point it to a system prompt (your AGENTS.md content) like this:

Code: Select all

{

"system": "See project documentation in ./AGENTS.md"

}

Or paste the content of AGENTS.md directly as the value of the "system" key if you want it self-contained. OpenCode reads this config at startup and uses it as the AI's base instructions for the session.

Pi (Inflection AI)
Pi is a conversational assistant — it doesn't have a CLI or file-loading mechanism. If you want Pi to understand FFastrans, open AGENTS.md and paste the relevant sections at the start of your chat. Pi will hold that context for the rest of the conversation.