本文へ移動
cccskills
無料GitHub で公開

timebox

Run an authorized task inside an Available Work Time (AWT) window with a shorter Closeout Grace Period (CGP), fixed deadlines, forecast checks, proportional convergence points, and a hard stop. Use when the user explicitly requests timeboxing, supplies an AWT/CGP pair, says AWT or CGP, or an applicable project policy requires the protocol. Do not invent durations or activate timeboxing merely because a task may take time.

インストール方法を見る

含まれるファイル(3)

  • SKILL.md7.6 KB
  • agents/openai.yaml201 B
  • references/independent-monitor.md4.7 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Timebox: AWT/CGP Execution

Timeboxing is an outer time and convergence envelope around authorized work. It does not add scope or replace the task's acceptance criteria.

AWT is the primary Available Work Time. Plan substantive work, verification, organization, and delivery inside it. CGP is a shorter Closeout Grace Period for an already-converged deliverable, an unexpectedly delayed final check, narrow closeout organization, or communicating the result. CGP is never a second work window.

Activate only with authority

Use this protocol only when one of these conditions supplies authority:

  • The requester explicitly asks for timeboxing.
  • The requester supplies an AWT/CGP pair or names AWT or CGP.
  • An applicable project policy requires the protocol.

Never infer a timebox from task size, urgency, or an estimate. Never invent a duration.

Establish the fixed clock

Accept any unambiguous duration units. Validate the supplied pair before doing timeboxed work:

  • AWT > 0
  • 0 <= CGP < AWT
  • CGP is materially shorter than AWT and sized for closeout, not ordinary work.

When the context already establishes AWT/CGP, interpret shorthand such as 30/10 as AWT 30 minutes and CGP 10 minutes. Do not create a confirmation loop for clear shorthand. If either duration is missing, ambiguous, or inconsistent with the two roles, ask the requester. Do not silently repair the pair.

Record a timezone-bearing original start and calculate:

normal_deadline = original_start + AWT
hard_stop       = normal_deadline + CGP

The original clock never resets after a resume, compaction, monitor creation, tool delay, or other interruption. If timeboxing begins after work has started, use the actual start when it is established. Ask if the start is unresolved and would materially change the deadlines.

Pass the hard preflight gate

Every new timezone-bearing original_start requires a fresh, visible backward plan. This applies when starting a new task, continuing under a requester-authorized new AWT/CGP pair, or reusing an existing monitor. A prior timebox's plan never carries forward.

After reading applicable instructions and doing only the minimum read-only orientation needed to identify the task and monitor, emit one concise commentary update containing all of the following:

  1. the timezone-bearing original_start, calculated normal_deadline, and calculated hard_stop;
  2. the requested outcome and smallest acceptance evidence;
  3. the critical path, required work, and explicit exclusions; and
  4. a backward schedule from normal_deadline through the 11/12, 75%, 50%, and 25% convergence points, naming the latest acceptable tangible state at each point.

Emit this update before sending or rebinding an independent monitor and before task-specific research, dependency resolution, implementation, builds, tests, or mutation. Hidden reasoning, a forward task list, or prose that merely names the next action does not satisfy the gate. If any field is absent, complete the visible preflight before dispatching the monitor or starting execution.

After the preflight is visible, bind the monitor once for this timebox when one is authorized, then execute immediately from the critical path and backward schedule. Do not wait for monitor setup or acknowledgment. Plan substantive work, verification, organization, and delivery inside AWT, with no ordinary work in CGP.

This skill works alone. If the MSW skill is also installed or otherwise required, use MSW to decide what work is necessary and this skill to govern how that necessary work converges inside the authorized clock.

If the CODER Loop is active, place one fixed clock around the complete loop and keep the CODER coordinator as clock owner. Workers and reviewers receive relevant fixed boundaries but never start replacement clocks. At the hard stop, CODER withholds acceptance for every unsettled claim and reports the strongest proven candidate.

If Codex Voice Optimizer controls the work, send it established material clock state for relay. CVO never calculates elapsed time, decides that a boundary was crossed, resets the clock, activates an extension, or becomes the independent monitor.

If MSL is active, apply it only to user-facing Timebox updates after the working task has calculated the clock state.

Work and converge

Check the completion forecast before every major phase and at least every min(15 minutes, AWT / 8), adapting the check to very short work windows. A forecast should answer only what changes execution:

  • What required result and proof are complete?
  • What required work remains?
  • When will the task finish at the current trajectory?
  • What is the leading risk to the normal deadline?
  • What one adjustment is necessary now?

Use these latest-acceptable convergence points:

  • At 25% of AWT, the contract, evidence, critical path, and exclusions are clear, and substantive work is underway.
  • At 50%, recalculate from tangible progress and correct scope or sequence.
  • At 75%, a tangible core deliverable exists, and speculative expansion has ended.
  • At 11/12 of AWT, required work is converged, and final verification and delivery preparation dominate.
  • At 100% of AWT, begin no new work.
  • At AWT plus CGP, stop and report the strongest available result.

When the forecast misses the normal deadline, protect the contract, end optional depth, simplify the route, and change the sequence immediately. Time pressure does not authorize dropping a required acceptance criterion.

Keep forecast checks quiet unless they reveal a material correction, crossed boundary, completion gap, or hard-stop risk. They exist to change decisions, not to create status ceremony.

Close out and stop

Enter CGP only for:

  • an already-converged deliverable;
  • an unexpectedly delayed final check;
  • narrow closeout organization; or
  • communicating the result.

During CGP, do not begin investigation, redesign, broad remediation, or other ordinary work. Finish early when the contract is complete and proven. Do not pad the task because time remains.

At the hard stop, stop all work and distinguish the result exactly:

  • complete and verified;
  • complete but unverified;
  • incomplete; or
  • unproven.

Report the outcome against the contract, the available proof, and any required work still open. Never imply completion that the evidence does not support.

Optional independent monitor

The normal experience is self-monitoring inside the working agent. Do not simulate an external timer with shell sleeps, polling loops, daemons, browser control, desktop control, repeated prompt injection, or a lifecycle hook.

Independent monitoring was designed first for the Codex app, whose separate tasks, task reading and messaging, and same-task heartbeat provide the intended topology. The same design can work in any harness that has an isolated observer, a way to inspect the working task, a reliable recurring wake mechanism, and a way to message the working task.

When the host has those capabilities and an independent monitor is authorized, read references/independent-monitor.md. Otherwise, use the self-monitoring workflow above and state once that independent monitoring is unavailable.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Run the Codex Optimized Development, Evaluation, and Remediation (CODER) Loop with an orchestration-only coordinator, non-overlapping task-family owners, fresh independent reviewers, evidence-scoped remediation, and final coordinator acceptance. Discover and compose optional MSW, Timebox, MSL, and Codex Voice Optimizer companions as the Slopware Dev Stack without making them dependencies. Use in Codex when the user invokes the CODER Loop, asks to divide a complex change into bounded implementation families with independent reviewers, requests multi-agent implementation without overlapping ownership, requires review and repair until every blocking gap is independently rechecked, asks for the CODER tutorial or Slopware Dev Stack, or wants CODER to install or use its companions. Do not trigger for an ordinary single-agent change, a review-only request, or a non-Codex host.

日本語の概要は準備中です。原文の説明を表示しています。

transcendr/slopware-skills362026年10月11日 更新

Create Codex threads and send messages into Codex desktop-app (GUI) threads, then wait for and read their replies, so Claude can orchestrate Codex agents. Use when the user asks to message, direct, steer, start, check on, or hand work to a Codex thread or agent (often given as a thread ID like 01a10323-9edd-...), or to run Codex as a sub-agent.

日本語の概要は準備中です。原文の説明を表示しています。

transcendr/slopware-skills362026年10月11日 更新

Optimize ChatGPT Voice in the Codex desktop app into an ear-first control plane for free-form task coordination and opt-in workflows. Apply spoken synthesis, routing-only coordination, owning-task role contracts, project placement, explicit authority, safe speech, current-state verification, honest recovery, and the optional Slopware Dev Stack: MSL, MSW, CODER Loop, and Timebox. Load named workflows only when invoked: Freeway for parallel lanes, Decision Walkthrough for read-only authority convergence, and Tutorial for teaching-first onboarding. Use when the user invokes the voice optimizer, coordinates Codex work through voice, assigns the current voice task as coordinator, runs Freeway, starts a decision walkthrough, learns voice orchestration, routes development through CODER, authorizes Timebox, or asks CVO to install or use Slopware companions.

日本語の概要は準備中です。原文の説明を表示しています。

transcendr/slopware-skills362026年10月11日 更新

msl

無料

Apply the Minimum Sufficient Language (MSL) principle through the MSL Kernel to write anything a reader must act on. Bind the reader and what they already know, partition facts from the machinery that produced them, emit each admitted fact as an action, a verification, a judgment, or a state, and stop when every claim is checkable and the reader can act. Use for status updates, work-in-progress reports to your own user, closeout and convergence comments, tickets, pull request and review write-ups, incident and postmortem notes, emails, memos, and any draft for readers outside the work. Also use when the user says MSL, Minimum Sufficient Language, write this for stakeholders, drop the jargon, make this readable, or asks for a paste-ready draft.

日本語の概要は準備中です。原文の説明を表示しています。

transcendr/slopware-skills362026年10月11日 更新

msw

無料

Apply the Minimum Sufficient Work (MSW) principle through the MSW Kernel to scope, execute, verify, and stop agent work. State the requested outcome and smallest proof, admit a claim only when deleting it would leave the contract unmet or unproven, do and prove each necessary claim, and halt at the fixed point. Use for implementation, debugging, review, research, planning, documentation, or any task where scope control, proportional evidence, or avoiding overwork matters. Also use when the user says MSW, Minimum Sufficient Work, do only what is necessary, avoid over-engineering, or stop when the result is proven.

日本語の概要は準備中です。原文の説明を表示しています。

transcendr/slopware-skills362026年10月11日 更新

transcendr のスキルをすべて見る

このスキルの問題を報告する