Terminal Context with $FORGE_TERM

When you run : fix this after a failed command, ForgeCode has no idea what you just ran. It can't see your last cargo build output or know that npm test exited with code 1. You end up narrating your terminal back to an agent sitting right next to it. That's backwards.

FORGE_TERM is on by default. The zsh plugin tracks the commands you run — what they were, whether they succeeded, and when — and passes that history to ForgeCode every time you invoke :. The agent knows what you ran and what failed without you explaining any of it.

This changes the interaction from:

# you have to reconstruct what already happened
: the build failed with a type error in src/main.rs on line 42 because...

to:

cargo build  # fails
: fix it     # ForgeCode already knows what failed

Disabling It

Context capture is on by default. To turn it off for the current session:

export FORGE_TERM=false

To disable it permanently, add to ~/.env (ForgeCode-only) or ~/.zshrc (shell-wide):

FORGE_TERM=false

Reload after editing: source ~/.zshrc

To re-enable:

unset FORGE_TERM

Verifying It Works

Check the variable is set:

echo $FORGE_TERM

Then run a command and ask ForgeCode about it:

cargo build
: what just failed?

If ForgeCode references the command you just ran, context capture is working.

Controlling Buffer Size

FORGE_TERM_MAX_COMMANDS sets how many commands the plugin keeps in the buffer. The default is 5.

export FORGE_TERM_MAX_COMMANDS=20

A larger buffer gives ForgeCode more history to work with, at the cost of a larger context window. If your prompts are hitting model context limits, lower it. If you work in long pipelines where ForgeCode needs to see further back, raise it.

What Comes Next

Once FORGE_TERM is set, : fix it means exactly what it says — ForgeCode knows the last thing that broke without you narrating it. The rest of the zsh integration — agent switching, file tagging, conversation management — is covered in ZSH Support.