Command Code sessions in your Mac notch

Taste-led coding, with every local session still in view.

Download for macOSFree 3-day trial · Apple silicon
AgentPeek monitoring Command Code sessions and local usage on a Mac
Sessions
GUI, terminal, and v3 history
Permissions
Terminal handoff only
Direct Chat
Saved session, no live owner
Usage
Monthly tokens, daily history
Installs
~/.commandcode/settings.json

Terminal and desktop sessions, one identity

AgentPeek recognizes cmd, cmdc, command-code, and commandcode, plus the signed Command Code desktop app. Active terminal or GUI work appears in the notch, while local records in ~/.commandcode/projects keep recent sessions available after the process closes.

Install the current CLI with npm i -g command-code@latest, authenticate with cmd login, then start it with cmd. Command Code remains the app that runs the agent; AgentPeek is its local monitor and control surface.

Stable hooks observe without taking permission control

AgentPeek adds bounded observer commands for SessionStart, PreToolUse, PostToolUse, and Stop under Command Code's existing settings groups. Unrelated settings and hooks are preserved.

Command Code checks its own native permission policy before PreToolUse, so AgentPeek never presents that observer as an approval or waiting-prompt channel. Native permissions stay in Command Code; the session's Open Terminal action returns you to the owning surface. Plan mode skips hooks entirely, so process and local-session discovery remain the honest fallback instead of inventing lifecycle events.

Local conversation and usage from the v3 record

Command Code's append-only v3 JSONL keeps a header and parent-linked entries. AgentPeek follows the active branch to show visible user and assistant messages, model, working directory, and per-session input, output, cache-read, and cache-write tokens. Unknown future entries stay unavailable instead of being guessed, and a torn final record does not erase the valid history before it.

The same local records feed calendar-month token totals and daily history. A persisted costUsd value is shown exactly when present. Command Code exposes no provider quota percentage or reset window there, and AgentPeek never reprices a record whose cost is absent.

Direct Chat only when no process owns the session

A session with a saved ID can continue through cmd -p --resume <id> --output-format json -- <prompt>. At send time, AgentPeek enables that path only when no live Command Code process owns the record, avoiding two writers against the same append-only session.

There is no live-TUI injection and no ACP claim. Experimental Mods are not required. While a live Command Code process owns the session, it stays in its own terminal or desktop app. Once that live owner exits, the proven headless resume path can become available.

Quick Routes to the files Command Code owns

Quick Routes opens shared and Command Code skills, agents, commands, instructions, MCP, config, Mods, taste, plans, projects, file history, logs, and ~/.commandcode. These are global, home-relative destinations. The Mods folder is a route to user-owned files, not a dependency on Command Code's experimental Mods API.

Questions

Yes. AgentPeek recognizes the Command Code CLI and desktop app, observes stable lifecycle hooks, reads local v3 conversations and usage, and continues eligible saved sessions after their live owner exits through the headless resume command.
No. Command Code evaluates native permissions before its PreToolUse observer, so AgentPeek cannot turn that hook into a waiting prompt or decision. The session remains visible and its Open Terminal action returns to Command Code.
Only for a saved session with no live Command Code process owning it at send time. AgentPeek uses cmd headless resume with JSON output; it never injects text into the live TUI and does not claim ACP support.
Calendar-month input and output tokens, cache breakdowns, daily history, and exact persisted cost when Command Code records it. AgentPeek does not infer a provider quota, reset window, or missing cost.
No. AgentPeek uses stable hooks, process discovery, and local session records. Quick Routes can open the Mods folder, but the experimental Mods API is not required for monitoring or Direct Chat.

Keep your agents in view

Download for macOSFree 3-day trial · Apple silicon