vincentkoc/dotskills/tmux-agent-lane-orchestrator

Claude Code and agent skill by vincentkoc/dotskills

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Monitor and coordinate one tmux agent lane, reconstruct worker state from panes and recent Codex logs, classify progress and blockers, and produce concise manager summaries. Use when multiple coding-agent workers run in tmux windows named L1, L2, and similar lane identifiers.

Install

npx vskill@latest install vincentkoc/dotskills/tmux-agent-lane-orchestrator

SKILL.md

tmux Agent Lane Orchestrator

Purpose

Turn a tmux window of coding-agent workers into a visible, auditable lane with clear state and next actions.

When to use

  • Monitor or summarize a tmux worker lane.
  • Reconstruct worker state after a manager cold start.
  • Detect duplicated, idle, waiting, blocked, or risky work.
  • Coordinate several coding-agent panes without mutating them blindly.

Workflow

  1. Establish lane scope.
  • Prefer an explicit lane number.
  • Otherwise infer it from a current tmux window named L<number>.
  • Do not inspect other lanes unless the operator expands scope.
  1. Capture lane state.
  • Resolve this loaded skill's directory, then run

python3 <skill-root>/scripts/lane_snapshot.py --lane <number>. Do not resolve the helper from the worker repository cwd.

  • Add --session <name> outside the active tmux session.
  • Linux inference requires a native Codex executable and bounded, structured

argv matching codex resume UUID or codex exec resume UUID exactly. Other argument forms stay unknown. Flat ps text cannot establish identity.

  • macOS reads executable paths only. Without a separately verified binding,

it reports unknown and the next action. Pass --thread L<number>.<pane>=<thread-id> only from independent evidence. A declaration still requires an unambiguous descendant Codex PID.

  • Add --json for machine-readable output.
  • Add --cursor-file <path> only when bounded incremental state is wanted;

the helper writes no persistent state by default.

  • Records larger than the smaller of --log-bytes and --total-log-bytes

are skipped through their next newline. The cursor reports discarding_record while skipping; later events resume.

  1. Cross-check panes and logs.
  • Pane titles alone are weak evidence.
  • Resolve pane ID to shell PID, descendant Codex PID, exact thread ID, the

state-database rollout path, and the newest turn.

  • Shared cwd and generic terms such as CI, failed, or running never

establish ownership.

  • Missing or conflicting identity is unknown, never a restore target.
  1. Classify each pane.
  • Use the newest exact turn and changing event/tool/token counters.
  • A completed turn after an earlier error is completed, not blocked.
  • Report bounded bytes read and truncation with the state.
  • Include the evidence and next action, not only the label.
  1. Intervene conservatively.
  • Inspect before steering.
  • Avoid duplicate heavy checks.
  • Do not kill or mutate another pane without explicit scope.
  • Prefer reversible actions and targeted commands.
  1. Summarize in manager style.
  • One line per pane: mission, state, evidence, blocker, next action.
  • Add one short manager judgment about where attention belongs.

Read references/factory-model.md when designing lane responsibilities or escalation policy.

Inputs

  • tmux session name.
  • Lane number or L<number> window.
  • Optional capture depth and per-rollout/total log byte limits.
  • Optional exact pane/thread declarations and bounded cursor path.
  • Optional operator-provided worker mission map.

Outputs

  • Current pane ID, shell PID, descendant agent PID, exact thread, and rollout snapshot.
  • Per-pane state classification with evidence.
  • Per-file and total log bytes read, cursor mode, and truncation status.
  • Concise manager summary and safe next actions.

Flow

mermaid
stateDiagram-v2
[*] --> ResolveOneLane
ResolveOneLane --> CaptureAndMatchIdentity
CaptureAndMatchIdentity --> ReportUnknown: missing or conflicting identity
CaptureAndMatchIdentity --> ClassifyNewestTurn: identity matched
ClassifyNewestTurn --> ReportLane: observation only
ClassifyNewestTurn --> TargetedIntervention: exact scope authorized
TargetedIntervention --> VerifyAndReport
ReportUnknown --> [*]
ReportLane --> [*]
VerifyAndReport --> [*]

Security

  • Security scanPassed, 95 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
0

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/vincentkoc/dotskills/tmux-agent-lane-orchestrator/badge)](https://verified-skill.com/skills/vincentkoc/dotskills/tmux-agent-lane-orchestrator)

Something wrong with this skill or its page? Tell us.