agentleFS
shawnclyborshawnclybor/clybor-claude-tooling/.claude/skills/multi-session/SKILL.md

multi-session

Skill6 starsChanged 28 days agoMITView on GitHub

Run, type into, and SEE other interactive Claude Code sessions from this one, through Orca terminal tabs. Use to test anything that only shows up on screen (the /workflows progress view, a slash command's output, a permission prompt), to run a workflow stage in a fresh session, or to run independent sessions side by side and read their results. Triggers — "open another session", "run it in a fresh session", "test what the screen looks like", "watch the other session", "drive Claude in a terminal". Needs the `orca` CLI. For a result you only need as text, use headless `claude -p` instead (cheaper, no screen). General Orca terminal and worktree reference: `orca skills get orca-cli`; supervised worker fleets: `orca skills get orchestration`.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

Files
SKILL.md

Multi-session — drive other Claude Code sessions

You are not limited to this session. From Bash you can open a new Claude Code session in an Orca terminal tab, type into it, and read its rendered screen. The user sees the tab in Orca too.

This skill covers what is specific to driving Claude Code. For everything else about Orca terminals (handles, receipts, worktrees), run orca skills get orca-cli.

Pick the lighter tool first

NeedUse
A result as text (an answer, a file written, a smoke test)claude -p "<prompt>" from Bash. No screen, cheapest
What a screen looks like, or an interactive-only featureThis skill
Many workers with task tracking and repliesorca skills get orchestration

The loop

# 1. Open. Launch Claude bare: a prompt on the command line can be lost (gotcha 1).
H=$(orca terminal create --worktree active --title "<name>" --json \
  --command "cd <dir> && claude --permission-mode acceptEdits --allowedTools 'Workflow,Read,Grep,Glob,Write,Edit,Bash'" \
  | grep -o 'term_[0-9a-f-]*' | head -1)
orca terminal wait --terminal $H --for tui-idle --timeout-ms 30000 --json   # read wait.satisfied

# 2. Look before you type.
orca terminal read --terminal $H --screen

# 3. Type. --wait-submit proves the turn started; input_accepted alone does not.
orca terminal send --terminal $H --text '<prompt or /command>' --enter --wait-submit 15

# 4. Watch. Foreground sleep is blocked: use Monitor with an until-loop on the screen.
until orca terminal read --terminal $H --screen | grep -q '<done marker>'; do sleep 5; done

# 5. Close when finished. This kills that Claude process.
orca terminal close --terminal $H

Keys: arrow down is --text $'\e[B', Esc is --text $'\e', and Enter alone is --text "" --enter.

Gotchas (all measured)

  1. The folder-trust dialog eats the first prompt. In a folder Claude hasn't opened before, a prompt passed on the command line is lost behind "Do you trust this folder?". Launch bare, read the screen, accept (arrow down, then Enter), then send the prompt.
  2. An open panel eats typed prompts. While /workflows (or any full-screen panel) is open, a sent prompt is dropped and the receipt still says input_accepted. Send Esc first. Treat a receipt without turn_started as not delivered.
  3. Use read --screen for anything visual. A plain read returns the raw stream: repainted lines arrive as stacked fragments and TUI layout is lost.
  4. /workflows opens only when sent on its own. Send it, then read --screen. With one run in the session it opens that run's progress view directly.
  5. Permissions decide whether a run stalls. An unattended session stuck on a prompt waits forever. Pass --permission-mode and --allowedTools at launch. A saved workflow also needs Workflow (or Workflow(<name>)) allowed, or it asks first.
  6. Each session costs a full context. A fresh session loads the CLAUDE.md files, skills and tools of the folder it opens in. Use a scratch folder with a tiny fixture when you only need to test behaviour.
  7. Close what you open. Tabs persist after your turn ends. List them with orca terminal list.

Report

Tell the user which tab you opened (its title), what you sent, and what the screen showed, quoting the lines from read --screen. Never describe a screen you did not read.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.