control-room
Skill2 starsChanged 2 days agoMITView on GitHub
Give the owner a live "project control room" page for multi-agent work run with the Workflow tool - KPIs, a stage stepper, the team of roles, every agent with status, model, time and result, the environment's health, and a reader for the Markdown deliverables, in light and dark. One self-contained HTML file, rebuilt every few seconds from the session's workflow journals; no server. Use whenever you start a multi-stage or multi-run Workflow project (a product build, a migration, an audit, an agent "team" with PM/CEO/dev/QA roles), when the owner asks for a dashboard, dashboard.html, a control room or a way to watch the agents, or asks "where can I see progress?". Also use it to publish a snapshot of the page as an Artifact in a cloud session.
In this repository9 other files it gives its agents.
- dsa-conscious-codingSkill
- excalidraw-architectureSkill
- improve-codebase-architectureSkill
- issue-to-phasesSkill
- pr-screenshotsSkill
- raise-prSkill
- visual-reviewerSkill
- wayfinderSkill
- workSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 2 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 17
use the same short phase titles in every script (phase('Build'), {phase: 'QA'}), one per stage of the plan.
- Very specific: line 18
add retries as "<label> (retry 2)".
- Very specific: line 30
Every key is described in references/config.md.
- Distinctive: line 61
When the owner says the page shows something wrong ("why is it still in progress?"), fix it in the config or the script.
- Distinctive: line 65
When the project ends or the owner pauses it, stop the background loop (TaskStop) and tell the owner that the page is now a static record.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Control room
A live page for the owner while a team of agents works. It reads what the Workflow tool already writes (one journal per run, one transcript per agent), so the work itself needs no changes. It is read-only: it writes one HTML file and nothing else.
S = this skill's base directory (the Skill tool prints it when the skill loads). Use the absolute path in commands.
1. Name things so the page can group them
The page groups agents by stage (the workflow's phase) and by role (a prefix of the agent's label). Before you write the workflow scripts:
- give every agent a label
"<Role>: <task>", e.g."QA: security audit","Senior dev: T-12 queue API"; - use the same short phase titles in every script (
phase('Build'),{phase: 'QA'}), one per stage of the plan; - add retries as
"<label> (retry 2)"; the page folds them into one row with an "attempt N" tag.
2. Write the config (optional)
Without a config the page still works: the stages are the phases in the order they first appear, and the roles are the label prefixes. For a project the owner will watch for hours, write control-room.json next to the project (in its tools/ folder or root) from assets/control-room.json.example. It sets:
- title and subtitle: the project's name and one line on what it is;
- stages: the plan's stages in order. Each stage lists the phase titles that belong to it, plus a one-line detail;
- roles: name, model and one line each, plus regexes on the label;
- environment: a health URL, the docker compose project, and links;
- docs: which Markdown files to show.
Every key is described in references/config.md. Never put a password, token or key in it: the page shows it.
3. Start it
Local session: run it in the background (Bash run_in_background: true) and leave it running for the whole project:
python3 $S/scripts/control_room.py --config <project>/tools/control-room.json --watch 10Its first output line is the HTML path. Give it to the owner as a clickable path; they open it in a browser and it reloads itself (Pause and Theme buttons in the header). It reads the session with the newest workflow run in this folder's Claude Code project. Pass --session <id> to pin this session (needed when other sessions run in the same folder), or --all-sessions to merge every session. --once --print writes once and prints a one-line summary: use it to check the page without a browser.
Cloud or remote session (the owner cannot open your files): write a static snapshot and publish it with the Artifact tool (icon chart), then republish the same path after each stage:
python3 $S/scripts/control_room.py --config control-room.json --refresh 0 --out control-room.htmlThe page contains the agents' results and the deliverables, which can name local paths and code. Artifacts are private by default; read the page before the owner shares the link.
4. Keep it honest
The page shows what the journals say. When reality differs, say so in the config, never by editing journals:
- An agent that wrote nothing for
stale_after_seconds(default 20 min) is shown as Failed with "Ended without a result": usage limits, network drops and stopped runs leave no journal entry. - A run you stopped and re-ran: add it to
supersededwith one line on why; its in-flight agents show as failed. - A task whose agent died but whose work another agent finished and you verified: add the label to
coveredwith the evidence. It shows as "Covered later". Never use this for work you have not checked. - A resumed or follow-up run that re-does a task: nothing to do. Only the newest attempt of each label is listed.
- A run whose script name the page cannot find: name it in
run_names. - Agents run outside workflows (a background Agent call): list them in the
extra_agentsJSON file.
When the owner says the page shows something wrong ("why is it still in progress?"), fix it in the config or the script. Explaining the error is not enough.
5. Stop it
When the project ends or the owner pauses it, stop the background loop (TaskStop) and tell the owner that the page is now a static record.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

