本文へ移動
cccskills
無料GitHub で公開

focus-receipt-proof-gate

Audit whether a "current product focus receipt" (Decision/Now/Why-now/Evidence/Not-now/Cut-suspend/First-visible-proof/ Acceptance-gate/Kill-revisit-trigger/Owner/Review-date) and the work order it gates state a real, testable decision and a real agent launch — not planning-placeholder prose that sounds like an assignment but can't be proven. Use when deciding whether to launch an agent chain against a milestone or a stated "focus," when a work order lacks explicit input/output/owner/proof-gate, or when an acceptance gate would only prove a cached UI state instead of daemon truth. NOT for distinguishing roadmap items from energy-driven sidequests (legible-roadmap-with-sidequests), reviewing a PRD/prototype for product-completeness gaps like cold start or pricing (product-reality-reviewer), or accepting already-finished work against a personal acceptance bar (port-daddy-user-surrogate-pm-review).

インストール方法を見る

含まれるファイル(12)

  • SKILL.md11.2 KB
  • agents/openai.yaml961 B
  • CHANGELOG.md185 B
  • examples/expected-output.md7.9 KB
  • examples/sample-input.json2.7 KB
  • examples/weak-input.json549 B
  • README.md1.5 KB
  • references/focus-receipt-schema.md5.4 KB
  • references/proof-gate-vs-cached-state.md4.4 KB
  • schemas/focus-receipt.schema.json5.0 KB
  • scripts/focus_receipt_audit.mjs10.8 KB
  • templates/output-template.md2.6 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。

Focus Receipt & Proof Gate

Decide whether a stated "focus" is a real, testable decision with a real agent launch behind it — or a planning placeholder dressed up as one.

Use This For

  • Auditing a milestone's "current product focus receipt" before agents are launched against it.
  • Checking a work order states real input, output, owner, and proof gate before calling it an agent launch.
  • Catching an acceptance gate that would only prove a cached UI state instead of daemon/event truth after a restart.
  • Verifying a focus receipt names an exit condition (Kill/Revisit Trigger) as explicitly as it names its entry condition (First Visible Proof).
  • Gatekeeping the transition from binder prose to agent work order at any milestone boundary.

Do Not Use This For

  • Deciding whether the wider roadmap should treat an item as planned work or an energy-driven sidequest (legible-roadmap-with-sidequests).
  • Reviewing a PRD, prototype, or vibe-coded concept for missing product fundamentals like cold start or pricing (product-reality-reviewer).
  • Accepting or blocking already-finished work against a personal acceptance bar (port-daddy-user-surrogate-pm-review).

Decision Model

flowchart TD
  A[Draft Decision / Now / Why-now / Evidence] --> B{Names First Visible Proof AND Kill/Revisit Trigger?}
  B -->|No| X[Reject: planning placeholder]
  B -->|Yes| C{Acceptance gate testable against daemon truth?}
  C -->|No, proves only cached UI state| X
  C -->|Yes| D[Draft work order: input / output / owner / proof gate]
  D --> E{All four stated?}
  E -->|No| X
  E -->|Yes| F[Launch the agent against the work order]
  F --> G[Re-check the proof gate from daemon truth after the work lands]
  1. Write the Decision/Now/Why-now/Evidence quartet before naming any agent. now is the narrow slice, not the whole milestone; whyNow must beat every other candidate; evidence is an observed fact, not an assertion — see references/focus-receipt-schema.md for the full field-by-field contract from the binder's build-prescription chapter.
  2. Name the First Visible Proof. The exact observable artifact — a screen, a saved event, a passing probe — that proves the decision produced something real, not a vibe.
  3. Name the Kill/Revisit Trigger. The exact condition under which the decision pauses and gets re-litigated. A receipt with an entry condition and no exit condition is half a decision.
  4. Write the acceptance gate as a claim rebuildable from daemon/event truth, per the binder's own rule that a proof "must survive relaunch from daemon truth, not from a cached UI model." Mark testableAgainstDaemonTruth: true only when that is actually possible — see references/proof-gate-vs-cached-state.md.
  5. Convert the receipt into a work order stating input, output, owner, and proofGate explicitly. The binder's own gate: "if a proposed chain cannot state its input, output, owner, and proof gate, it is a planning placeholder, not an agent launch."
  6. Audit both together with scripts/focus_receipt_audit.mjs before spawning any agent against the receipt.
  7. Re-run the audit after the work lands. The proof gate must still hold from daemon truth, not from the receipt's original prose.

Output Contract

A launch-ready focus receipt carries:

  • receipt: decision, now, whyNow, evidence, firstVisibleProof, acceptanceGate.{statement,testableAgainstDaemonTruth}, killRevisitTrigger, owner, reviewDate — all stated, none implied.
  • workOrder: input, output, owner, proofGate — all stated for every chain the receipt launches.

Use scripts/focus_receipt_audit.mjs to audit a spec matching schemas/focus-receipt.schema.json and return { pass, score, findings, recommendations }.

Anti-Patterns

Decision Without Entry Or Exit Criteria

Novice: Writes a confident Decision/Now/Why-now, then leaves First Visible Proof and Kill/Revisit Trigger implicit — "we'll know it's working when we see it, and we'll know to stop when it's clearly not." Expert: A real focus receipt names the exact observable artifact that proves entry (First Visible Proof) and the exact condition that ends or re-litigates the decision (Kill/Revisit Trigger) in the receipt itself, not in someone's head. Detection: focus_receipt_audit.mjs fires no-first-visible-proof (critical) when receipt.firstVisibleProof is empty/absent, no-kill-trigger (critical) when receipt.killRevisitTrigger is empty/absent, and receipt-missing-required-field (critical) for any other missing required field (decision, now, whyNow, evidence, owner, reviewDate, or acceptanceGate.statement).

Acceptance Gate Tests The UI, Not The Daemon

Novice: Treats "the panel shows the agent as compliant" as proof — a claim a stale cache or an optimistic UI update can satisfy without anything real having happened on the backend. Expert: An acceptance gate must be restated as a claim that survives relaunch from daemon truth — rebuildable from backend/event state after a restart or reconnect, never from what is currently painted on screen. Detection: focus_receipt_audit.mjs fires acceptance-gate-not-daemon-testable (critical) whenever receipt.acceptanceGate.testableAgainstDaemonTruth is not exactly true (fails closed on missing, false, or any other value), and review-date-elapsed (high) when receipt.reviewDate has already passed without the receipt being revisited.

Planning Placeholder Masquerading As An Agent Launch

Novice: "Launch an agent for governance work" — no stated input, no stated output, no accountable owner, no proof gate. Sounds like an assignment; is actually a wish. Expert: Per the binder's own rule, a chain that cannot state its input, output, owner, and proof gate is a planning placeholder, not an agent launch — do not spawn it. Detection: focus_receipt_audit.mjs fires placeholder-not-launch (critical), naming exactly which of workOrder.input / workOrder.output / workOrder.owner / workOrder.proofGate is missing or empty.

References

FileLoad When
references/focus-receipt-schema.mdDrafting or reviewing a focus receipt's twelve fields, or checking a milestone chapter's Goal/Tasks/Gate structure and the Agent count rule.
references/proof-gate-vs-cached-state.mdWriting or trusting an acceptance gate, or deciding whether a proposed chain is a real work order or a planning placeholder.
examples/expected-output.mdNeed to see a placeholder-dressed-as-a-decision receipt audited, then the same receipt fixed and passing.
templates/output-template.mdNeed a fill-in-the-blank focus receipt and work order template.
schemas/focus-receipt.schema.jsonNeed to validate a focus-receipt JSON payload's structure before auditing it.
scripts/focus_receipt_audit.mjsNeed deterministic scoring of a focus receipt and its work order.
agents/openai.yamlNeed a subagent descriptor for delegated focus-receipt/proof-gate review.
<!-- BEGIN BUNDLE INDEX (auto: index_references.py) -->

Skill Bundle Index

Every file in this skill, and when to open it. Auto-generated; run scripts/index_references.py --fix.

root

  • CHANGELOG.md — Focus Receipt & Proof Gate — Changelog — - Initial skill creation - Core process defined - Reference files and deterministic focus_receipt_audit script added
  • README.md — Focus Receipt & Proof Gate — Audit whether a "current product focus receipt" and the work order it gates state a real decision with real entry/exit criteria, a daemon-te

agents/

examples/

references/

  • references/focus-receipt-schema.md — Focus Receipt Schema — Use this when drafting or reviewing a "current product focus receipt," or when checking a milestone chapter's Goal/Tasks/Gate structure befo
  • references/proof-gate-vs-cached-state.md — Proof Gate vs. Cached State — Use this when writing or trusting an acceptance gate, or when deciding whether a proposed agent chain is a real work order or a planning pla

schemas/

scripts/

templates/

<!-- END BUNDLE INDEX -->

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

Expert in 2000s-era music visualization (Milkdrop, AVS, Geiss) and modern WebGL implementations. Specializes in Butterchurn integration, Web Audio API AnalyserNode FFT data, GLSL shaders for audio-reactive visuals, and psychedelic generative art. Activate on "Milkdrop", "music visualization", "WebGL visualizer", "Butterchurn", "audio reactive", "FFT visualization", "spectrum analyzer". NOT for simple bar charts/waveforms (use basic canvas), video editing, or non-audio visuals.

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。

curiositech/port-daddy22026年10月12日 更新

Expert legal research agent for finding and scraping expungement data state by state. Knows authoritative sources, URL patterns, Firecrawl configuration, and 2026 legal landscape.

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。

curiositech/port-daddy22026年10月12日 更新

Expert in 3D computer vision labeling tools, workflows, and AI-assisted annotation for LiDAR, point clouds, and sensor fusion. Covers SAM4D/Point-SAM, human-in-the-loop architectures, and vertical-specific training strategies. Activate on '3D labeling', 'point cloud annotation', 'LiDAR labeling', 'SAM 3D', 'SAM4D', 'sensor fusion annotation', '3D bounding box', 'semantic segmentation point cloud'. NOT for 2D image labeling (use clip-aware-embeddings), general ML training (use ml-engineer), video annotation without 3D (use computer-vision-pipeline), or VLM prompt engineering (use prompt-engineer).

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。

curiositech/port-daddy22026年10月12日 更新

Apply crisis decision-making research to agent routing, uncertainty triage, and coordination failure analysis in time-pressured systems. Use when diagnosing handoff failures, analytical paralysis, or expert judgment under incomplete information. NOT for routine coding, simple CRUD design, or static single-agent tasks with complete information.

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。

curiositech/port-daddy22026年10月12日 更新

Use for insight, reframing, contradiction, impasse, and anomaly-driven problem solving when execution effort no longer helps. NOT for routine optimization, error correction, or well-specified tasks with known solution paths.

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。

curiositech/port-daddy22026年10月12日 更新

Apply cognitive task analysis to expert work that depends on perceptual cues, branching judgment, and recurring monitoring loops. Use when decomposing expert capability into agent structure, simulation design, or validation interviews. NOT for ordinary step-by-step SOP capture or simple pipelines with no tacit cue layer.

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。

curiositech/port-daddy22026年10月12日 更新

curiositech のスキルをすべて見る

このスキルの問題を報告する