跳到正文
cccskills
免费在 GitHub 公开

conserved-capacity-admission-and-settlement

Designs and audits native-unit capacity reservation, serial admission, ambiguity holds, stop reserves, effect closure, and control-disjoint settlement handoff for consequential agent attempts. Use when scarce capacity must be conserved across retry, rework, resurrection, or compensation. NOT for scheduling work, choosing models, pricing unlike resources into one scalar, authorizing effects, grading work, paying workers, or asserting runtime containment.

中文简介正在准备中。当前显示原文说明。

实测评分 未实测

查看安装方法

包含的文件(8)

  • SKILL.md7.5 KB
  • CHANGELOG.md728 B
  • examples/valid-no-settlement-attempt.json2.4 KB
  • references/authority-and-conservation.md3.3 KB
  • schemas/conserved-attempt-v1.schema.json5.6 KB
  • scripts/test-bundle.mjs3.0 KB
  • scripts/validate-conserved-attempt.mjs7.6 KB
  • tests/activation.md750 B

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Conserved Capacity, Admission, and Settlement

Use this skill to keep four authorities separate:

  1. a scheduler proposes an eligible attempt;
  2. a capacity broker reserves a native resource vector;
  3. an admission writer consumes both exactly once and creates a lease;
  4. an effect boundary closes authority and hands evidence to an independent settlement authority.

The skill designs or audits records. It cannot reserve real capacity, admit a body, issue a capability, schedule a node, settle compensation, or prove containment.

Activate when

  • a node attempt, retry, repair, renewal, consuming gather, or resurrection needs scarce capacity;
  • a crash or lost acknowledgement could duplicate a lease, effect, refund, slash, or payout;
  • an external effect may be ambiguous and capacity reuse must fail closed;
  • Stop must retain capacity for reconciliation and teardown;
  • effect closure must not wait for compensation or dispute settlement;
  • a no-settlement profile still needs explicit conservation and terminal records.

Do not activate for

  • deciding which roadmap node runs next;
  • estimating demand, price, or model quality;
  • turning tokens, subscription allowance, cash, time, concurrency, and attention into one invented currency;
  • deciding whether an effect is authorized;
  • grading whether work is correct;
  • making legal, tax, employment, custody, or moral-status decisions.

Authority and resource rules

  • Native vectors only. Every dimension keeps its native unit and source: calls, tokens, seconds, dollars, slots, attempts, network bytes, or attention events.
  • Reservation is not admission. ReservationCommit is exact, expiring, one-use capacity evidence. It cannot create a lease.
  • Admission has no scheduling discretion. The admission writer may consume an exact EligibilityCommit and ReservationCommit; it may not select a different node, route, provider, or resource vector.
  • Every consuming attempt pays capacity rent. Retry, renewal, rework, successor, elastic slot, provider call, and model-consuming gather each bind to a current reservation.
  • Stop reserve is protected. Reconciliation, fence, teardown, and receipt publication have explicit reserved capacity that ordinary work cannot consume.
  • Ambiguity holds resources. AMBIGUOUS and QUARANTINED_UNRESOLVED expose no retry authority and do not release affected capacity.
  • Containment does not wait for settlement. Effect closure and an idempotent settlement-outbox record share one local transaction. Settlement consumes asynchronously.
  • Settlement is control-disjoint. Capacity broker, scheduler, admission writer, effect boundary, and settlement authority are distinct principals and key domains.
  • Compensation is a new effect. Payment, refund, slash, reputation, and salvage funding require their own authorization and receipt.
  • No-settlement is explicit. It means no bond, bounty, payout, slash, or refund; model/API usage remains accounted in its own dimensions.

State spine

stateDiagram-v2
  [*] --> Eligible: exact EligibilityCommit
  Eligible --> Reserved: one-use ReservationCommit
  Reserved --> Admitted: atomic consume + lease + durable dispatch intent
  Admitted --> Running: observed start
  Running --> EffectClosed: effect authority closed
  Running --> Ambiguous: lost or conflicting acknowledgement
  Ambiguous --> Reconciled: independent evidence
  Ambiguous --> Quarantined: bounded workflow terminal
  Reconciled --> EffectClosed
  Quarantined --> Quarantined: capacity held; no retry
  EffectClosed --> HandoffReady: settlement outbox committed
  HandoffReady --> Settled: independent terminal
  HandoffReady --> Unsettled: disagreement or unavailable authority
  Settled --> [*]
  Unsettled --> [*]

Quarantined is a workflow terminal, not proof that the external effect did not occur. A separately reserved no-effect observer may investigate, but is neither retry nor successor.

Required attempt identity

Bind every record to:

  • plan digest and revision;
  • node ID, attempt ID, and body generation;
  • principal and authority epoch;
  • repository, immutable target, route, and provider profile;
  • policy, eligibility, reservation, review, and evaluation digests;
  • idempotency key and expiry.

A mismatch rejects the join. “Same task” or “same agent” is never enough.

Crash and conservation method

  1. Declare resource dimensions and their ledger owners.
  2. Define the reservation linearization point.
  3. Atomically write reservation consumption, lease, and durable dispatch intent.
  4. Crash immediately before and after every durable write.
  5. Duplicate, reorder, delay, and lose every acknowledgement.
  6. Reconstruct balances from receipts only.
  7. Prove no lease without consumed capacity, no double release, and no reuse under ambiguity.
  8. Close effect authority before asynchronous settlement.

Anti-patterns

Free subscription capacity

Wrong: record a subscription-backed model call as zero because no invoice line exists. Right: track remaining allowance, reset window, model tier, call/token estimates, and confidence in native units.

Broker as scheduler

Wrong: the capacity broker picks the next node because it sees availability. Right: it returns an exact reservation or denial for a scheduler-supplied attempt.

Ambiguous retry

Wrong: timeout releases capacity and retries automatically. Right: retain the hold and require independent reconciliation.

Settlement in Stop

Wrong: teardown waits for a payout or slash decision. Right: close effect authority, commit the handoff, then settle asynchronously.

Output and validation

node skills/conserved-capacity-admission-and-settlement/scripts/validate-conserved-attempt.mjs \
  skills/conserved-capacity-admission-and-settlement/examples/valid-no-settlement-attempt.json
node skills/conserved-capacity-admission-and-settlement/scripts/test-bundle.mjs

Static success proves only the supplied record. It never authorizes scheduling, execution, payment, retry, or release.

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

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.

中文简介正在准备中。当前显示原文说明。

curiositech/port-daddy32026年10月19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.

中文简介正在准备中。当前显示原文说明。

curiositech/port-daddy32026年10月19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).

中文简介正在准备中。当前显示原文说明。

curiositech/port-daddy32026年10月19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.

中文简介正在准备中。当前显示原文说明。

curiositech/port-daddy32026年10月19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.

中文简介正在准备中。当前显示原文说明。

curiositech/port-daddy32026年10月19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.

中文简介正在准备中。当前显示原文说明。

curiositech/port-daddy32026年10月19日 更新

查看 curiositech 的全部技能

报告此技能的问题