Skip to content

Latest commit

 

History

History
33 lines (20 loc) · 2.53 KB

monitoring.md

File metadata and controls

33 lines (20 loc) · 2.53 KB

Agent session monitoring

The image includes abtop, a terminal monitor you start by hand. It is pinned in versions.env (versions).

Contents

Running abtop

Run inside the container, in a separate terminal or tmux window:

abtop

Upstream documents session, token, process and port monitoring for Claude Code, Codex and OpenCode (supported agents). Antigravity is not listed. OpenCode monitoring needs sqlite3, which the image includes. Inside a container, abtop sees only that container's processes; it is not a host-wide dashboard.

The TUI can terminate processes. x targets the selected session, and X targets orphan ports. Press q to quit and t to cycle themes. The theme choice persists in ~/.config/abtop/config.toml on the home volume. abtop keeps its default theme and has no custom themes. Version 0.5.5 compiles its 12 themes in, and config.toml only selects one by name. The default background is #191919 (Color::Rgb(25, 25, 25) in src/theme.rs), so it does not match the GitHub Dark #010409 background. See upstream key bindings and configuration.

Model calls and controls

The source links below point at the pinned release, v0.5.5.

  • The TUI and abtop --once generate session titles through claude --print, sending excerpts of the first prompt and response. These calls can use Claude quota (summary implementation). Not verified: a supported switch that turns this off in the TUI.
  • abtop --json prints a one-shot snapshot without the summary calls (CLI source). Its output can contain transcript text and project paths, so keep it out of public logs.
  • Agentainer does not start abtop, run abtop --setup, install its rate-limit hook or replace an existing status line.

Verification

Verified by the offline container test: the installed version, parsing of live and demo JSON snapshots, and that those runs create no Claude settings. Not verified: detection and token accuracy for the pinned agents, rate-limit collection, interactive controls and cross-agent discovery.