Interactive mode

The IsonForge REPL is the default surface. Just run isonforge and start typing.

Input editing

Key	Action
`Enter`	Submit prompt
`Alt+Enter`	Insert newline (multi-line input)
`Shift+Enter`	Insert newline (terminals that distinguish - many don't; prefer `Alt+Enter`)
`Backspace`	Delete previous character
`Left` / `Right`	Move cursor
`Up` / `Down`	Navigate prompt history (within session)
`Ctrl+A`	Move cursor to start of current line
`Ctrl+E`	Move cursor to end of current line

The prompt persists per-line history within the session. Up arrow walks oldest-first into history; saves your current draft so Down arrow can restore it.

Mode and control

Key	Action
`Shift+Tab`	Cycle permission mode: default -> acceptEdits -> plan -> auto -> bypassPermissions -> default
`Ctrl+C`	Abort stream (if streaming) or exit at idle
`ESC`	Abort stream, or deny a pending permission prompt

The permission mode shows in the status line. Shift+Tab is fast - tap through modes mid-task to relax or tighten what the agent is allowed to do.

Inline shortcuts

These work at the prompt and are processed client-side before the agent runs.

`!command` - run bash directly

> !ls -la

Executes locally. Output streams to your terminal. The agent does not see it - this is for when you want to do something yourself without invoking the model.

`@path/to/file` - inject file content

> what's wrong with this function? @src/auth.py

IsonForge expands @src/auth.py to the file's content inline before sending to the agent. Files >100KB are skipped with a warning. Multiple @ references work in one prompt.

`#note` - append to global memory

> #the deploy script needs `--env=staging` flag

Saves the note to ~/.isonforge/global-memory.md with a timestamp + project name. Cross-session, cross-project. See Memory.

Slash commands

Type / to autocomplete. See Slash commands for the full list. Highlights:

/clear - reset conversation
/plan - propose-before-execute mode
/checkpoint + /rollback - safe file editing
/cost - token usage
/skills - list available Skills

Streaming UX

While the agent thinks, you'll see:

A spinner with status text (Thinking..., Planning..., tool name).
The live "thinking" pane (warm sage box) if reasoning mode is on. Shows the model's chain-of-thought as it streams.
Tool calls render as fenced boxes with the tool icon, name, and one-line argument preview.
File diffs render inline with +/- highlighting before each edit is applied (unless --yolo).

Permission prompts appear inline when the agent calls a tool that requires approval. Pick one with arrow keys + Enter:

Yes
Yes, always (this session) - adds to session allowlist
No - the model is told the call was denied and continues

For dangerous bash commands, "always" is removed and No is the default.

Multi-line prompts

For longer prompts, use Alt+Enter to insert newlines. Each line gets a continuation prefix (...). Plain Enter submits. Useful for pasting code snippets, error messages, or specs:

> fix this:

  TypeError: cannot read property 'id' of undefined
    at User.findById (auth/user.js:42:15)
    at /api/users/me (routes/users.js:18:10)

  the error fires on first request after restart, then never again

Resume + history across sessions

/sessions - list saved sessions in this project
/resume <id> - load one (REPL-internal)
isonforge --resume - interactive picker at startup
isonforge --continue - load the most recent
isonforge --continue <id> - load a specific session

Session files live at ~/.isonforge/sessions/<id>.json (mode 0600).

Title bar / status

The terminal title shows: isonforge - <project> - <model>. The status line at the bottom of the REPL shows the current permission mode, turn count, and token usage.

Exit cleanly

Ctrl+C once aborts a stream. Ctrl+C again (or while idle) exits. exit or quit also exit. Background agents and worktrees persist - your work isn't lost.