agent-shell-watch

Reviewed at commit adf5b83

A status line and a /shell-watch pane for every Bash call, background task and delegated agent run of the session (Codex, Pi, Devin), with elapsed time…

Made by
apolenkov
What we checked

Shows a status line and a side pane for Codex, Pi, Devin and OpenCodeReview runs started from Bash (or every Bash call and background task if you set scope to all): elapsed time, quiet or hung output, verdicts, exit codes. It reads those tools' local usage and limit files to show tokens and limits. It makes no network calls and changes no tool calls.

Review details

Static review of the source at the pinned commit; not run. Registers session.start (two handlers), tool.call, session.append, command.run (/shell-watch), ui.open, ui.close and ui.render for its Pane; every hook passes the event on unchanged. Runtime calls: $.ui.status, $.ui.open/close, $.command.register, $.clock.every (1 s tick, 2 s poll, 2 s limits check), $.agent.list, $.session.messages (rebuilds Bash calls from the local transcript), $.fs.stat/list/read, $.env.get HOME, $.store.get/set (only the view and whether the pane was open), and $.process.run with a fixed argv of tail -n 40 or tail -c on a path. Those paths come from Bash output (Output is being written to) or from a command's --watch-file, tee or redirect target, so a command the model ran can make it tail a file it named; the text is only shown in the local pane and goes nowhere else. For Codex, Pi and Devin it reads ~/.local/state/executor-limits/<name>, the newest ~/.codex/sessions rollout files of today and yesterday and ~/.pi/agent/sessions (end of file only), and extracts only rate-limit percentages, resets, token counts and cost; this is throttled and only runs while the runners view is open or a runner is live. The pane has a Stop button that calls the TaskStop tool for a background task through $.tool.call, only on a press. No fetch, sockets, eval, dynamic import, settings or permission changes. Changes since the previous review: a scope option (runners by default; all keeps every Bash call) and call-filtering helpers in track.ts, backfill.ts and calls.ts. Shell and CI files in the repository (scripts/*.sh, .github night-merge workflow) are development tooling and are not loaded by the plugin. Read in full: manifests, register.tsx, track.ts, slash-command.ts, pane.tsx, limits.ts, model/limits.ts, model/poll.ts, model/parse.ts, model/backfill.ts, model/config.ts. Read by targeted search and partial reading only: model/calls.ts, format.ts, groups.ts, layout.ts, outcome.ts, pane-items.ts, runners.ts, usage.ts (beginning), view/*.tsx (pure rendering and parsing). npm dev dependencies not reviewed.

What it uses
  • Reads your prompts: no
  • Sees tool calls: yes
  • Blocks or changes tool calls: no
  • Reads or writes files: yes
  • Runs programs on your machine: yes
  • Calls the model (uses tokens): no
  • Sends data over the network: no
  • Stores data: yes
  • Shows something on screen: yes
Hooked events

Always-on status line and a pane for this session's Bash calls, background tasks and Codex/Pi/Devin/OpenCodeReview runs: elapsed time, output freshness, verdicts and errors.

Made this mod? Add a badge to your README

modsforClaude badge for agent-shell-watch

Paste this into your README. The badge says reviewed once the current version has been reviewed, and links readers to this page.

[![Listed on modsforClaude](https://modsforclaude.com/badge/agent-shell-watch.svg)](https://modsforclaude.com/mods/agent-shell-watch/)

← All mods