DEV Community

Deduped
Deduped

Posted on Originally published at dedupedai.com

Claude Flightdeck: a live dashboard for Claude Code context, cost, permissions and subagents

When Claude Code fans out five subagents, you mostly see a spinner. How full is the context window? What did that run cost? Which tool calls were allowed silently, and which were denied? Flightdeck is a Claude Code mod that answers all of that in a live pane next to your transcript: a context gauge with compactions, cost and rate-limit resets, a timeline of advisor and architect consults, one cell per permission check, and your subagents as cards or swimlanes.

Two things make it easy to trust. Every number comes from a real session event, and it only watches: every hook passes its event on unchanged, so it never denies, rewrites or delays a tool call. Nothing leaves your machine.

Requirements

  • Claude Code 2.1.287 or later (mods are an early access feature). Check with claude --version.
  • A terminal at least 144 columns wide if you want the pane to open by itself. Narrower terminals work with the /flightdeck command.

Install in four commands

Inside Claude Code:

/plugin marketplace add scasella/claude-flightdeck
/plugin install flightdeck@claude-flightdeck
/reload-plugins
/flightdeck
Enter fullscreen mode Exit fullscreen mode

Or from your shell:

claude plugin marketplace add scasella/claude-flightdeck
claude plugin install flightdeck@claude-flightdeck
Enter fullscreen mode Exit fullscreen mode

To try it for one session without installing, load it from a clone:

git clone https://github.com/scasella/claude-flightdeck
claude --plugin-dir ./claude-flightdeck
Enter fullscreen mode Exit fullscreen mode

If the installer says config options are not set, ignore it. The defaults are fine.

What each panel tells you

  • main: model, effort, permission mode and request count, a context gauge with a compaction counter, session cost, the first two rate-limit windows with time to reset (when your plan reports them), and time to first token for the last request.
  • architect: advisor or architect consults on a timeline, whether one is running and how long the last took.
  • gate: one cell per permission check. Green means allowed without asking, blue means decided by the auto mode classifier or you, amber is pending, red is denied, and dim means it happened inside a subagent.
  • agents: each subagent's task, type, live context and output tokens, steps and a running clock. max_tokens shows in red. When cards no longer fit, they become swimlanes on one time axis.
  • receipt and log: the last turn's duration, agents, edits, errors and added cost, plus a running log of prompts, spawns, consults, edits and denials.

Panels with nothing to show take no room, so a simple session shows just the main box and the log.

Three workflows worth setting up

1. Watch context and cost while you work

Keep the pane docked in a fullscreen terminal. When the gauge climbs toward a compaction, you know to wrap up a task or start a fresh session before the model loses earlier detail. The compaction counter tells you how many times that already happened.

2. Audit every permission check

Focus the pane with ctrl+x tab, then press f, s or o to open the file, shell or other drill-down. Each shows the last five checks and their verdicts, with credentials masked. This is the fastest way to see what an agent tried to touch during an unattended run.

3. Follow subagents

Start a task that spawns subagents, then press 1, 2 and so on to expand a card or lane: its full task, its last three tool calls and the start of its answer. A red max_tokens is your cue that an agent ran out of room.

Useful commands and options

/flightdeck            open the pane
/flightdeck close      close it
/flightdeck reset      clear agents, checks, consults, log and the turn
/flightdeck layout auto|compact|wide|mini
Enter fullscreen mode Exit fullscreen mode

Options live in /config, or in your user settings.json under pluginConfigs["flightdeck@claude-flightdeck"].options. Note that project level .claude/settings.json entries are ignored without a warning. Handy ones: maxCards (1 to 6 cards before swimlanes), motion: off if the animation is distracting, palette: pastel if colours look wrong, and statusLine to mirror context, agents and denials into the status line.

Troubleshooting

  • No pane: confirm version 2.1.287+, run /reload-plugins then /flightdeck. Look for a dim transcript line starting with flightdeck: that names the failing hook.
  • VS Code extension shows nothing: one report saw hooks run but no pane. Run claude in VS Code's integrated terminal instead.
  • Counters look stale after an update: /flightdeck reset.

Honest caveats

Mods are early access and their API can change between Claude Code releases, so expect occasional breakage after updates. Consult "moments" are inferred and labelled as such. It pairs well with a status line tool like claude-hud rather than replacing it.

Use cases with copy-paste prompts

Each page has who it is for, the steps and a prompt you can paste into your coding agent.

Use case: Watch context, cost and rate limits during a session

Use case: Audit every permission check a session makes

Use case: Follow subagents as cards or swimlanes


Get one email every Monday with the week's hottest AI repos and exactly how to use them: subscribe free at dedupedai.com

Top comments (0)