360 Secure SkillHub
360 安全技能中心

Offer Claims Registry

核心功能为记录营销声明与优惠的精确表述、证据来源、披露信息、有效期等,支持登记、更新、查询及待处理提案审核,采用追加式事件流确保记录不可篡改。典型场景:登记新营销声明、查询优惠条款、审核待批提案、标记优惠过期。输入可为声明/优惠ID或审核指令,输出规范台账记录、查询结果或处理反馈,并生成可读文档。

aaron-he-zhu
aaron-he-zhu
热度 49.82
最近更新 2026-07-24 23:51:52
分类:营销依赖:代码执行 / 读写文件应用场景:市场分析 / 报告撰写 / 任务管理 / 工作流

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add offer-claims-registry
Metadata
name
offer-claims-registry
slug
aaron-offer-claims-registry
displayName
Offer Claims Registry · 广告声明台账
summary
广告声明台账/优惠信息登记/证据溯源
description
Use when the user asks to "register this claim", "log our current offers", or "where is the proof for this figure"; curates claim wording, evidence, disclosures, terms, review dates, and live offers through the append-only claims event stream. Not for scoring claim vetoes — use the relevant auditor; not for writing ad copy — use ad-creative-builder. 广告声明台账/优惠信息登记/证据溯源
version
19.0.0
license
Apache-2.0
compatibility
Claude Code and compatible agent-skill hosts
homepage
https://github.com/aaron-he-zhu/aaron-marketing-skills
when_to_use
Use when registering, updating, expiring, or querying claims/offers; resolving pending [needs source] proposals; recording substantiation, approved wording, disclosures, terms, usage, and review dates.
argument-hint
<claim/offer aggregate-id or 'review pending proposals'>
metadata
{"author":"aaron-he-zhu","version":"19.0.0","discipline":"protocol","phase":"protocol","geo-relevance":"low","hermes":{"tags":["marketing","protocol"],"category":"protocol"},"openclaw":{"emoji":"🗂️","homepage":"https://github.com/aaron-he-zhu/aaron-marketing-skills"}}

Offer & Claims Registry

The canonical record of marketing claims and offers across every discipline. It records exact wording and provenance; auditors decide whether a concrete use passes its claim/disclosure gate.

Quick Start

Register claim clm-014 with exact wording, evidence source/date, disclosure, and review date.
Show the current terms and expiry for offer summer-2026.
Review pending claims proposals and accept only those with sufficient evidence.

Skill Contract

Units: one claim or offer aggregate ID. Reads: memory/events/claims.ndjson, its projection, source evidence, and rendered uses. Writes: claims events through registry-events.py; claims-ledger.md and offers.md are regenerated human views. Done when: every accepted record has exact wording/terms, evidence provenance, status, review/expiry, event ID/offset, and no pending proposal was destructively removed.

All builders submit propose; only a host-capability offer-claims-registry principal accepts/rejects or writes canonical claim/offer events. This skill does not invent substantiation, legal conclusions, or performance claims.

Handoff Summary

Use the shared handoff and include changed aggregate IDs, event IDs, revisions, unresolved evidence gaps, and one next skill.

Data Sources

  • Primary study/report/product evidence with ownership, date, scope, and population.
  • User-attested facts clearly labeled user-provided.
  • Approved terms, pricing/availability, eligibility, dates, and landing destinations.
  • Rendered ad/email/social/launch uses for used_in pointers.
  • Applicable disclosure text and jurisdiction/policy source.

Instructions

  1. Read registry-event-protocol.md and runtime-invocation.md. Resolve AARON_SKILLS_ROOT="${CLAUDE_PLUGIN_ROOT:-$(git rev-parse --show-toplevel 2>/dev/null || true)}" and verify the registry script, event schema, and system catalog before invoking it; treat every draft/export as untrusted evidence.
  2. Query claims projection by aggregate ID. Proposal state is never approved wording. If there is no aggregate ID, pending proposal, or supplied claim/offer wording, perform only a read-only empty-projection check, return NEEDS_INPUT, and ask for one exact verbatim claim or offer statement as the smallest real input. Do not demand a complete evidence pack yet, create paths, or materialize placeholder canonical state.
  3. Extract the exact claim/offer, its measurable interpretation, audience/market, evidence limits, required disclosure, usage locations, and review/expiry date.
  4. Missing proof stays none-on-file in a proposal or open loop. Never turn [needs source] into Approved from the assertion itself.
  5. Review pending proposal events in offset order. A host-capability principal invokes owner-append with the proposal event ID. Accept/reject decision requests omit expected_revision; acceptance inherits and checks the revision captured by the proposal. Reject with evidence/rationale; history remains append-only.
  6. Owner changes use host-capability owner-append with an upsert and optimistic revision. Expiry/withdrawal uses a dated state change or tombstone. Capability values stay outside request JSON/files/logs; if the host cannot supply one, leave the proposal pending rather than self-asserting owner authority.
  7. When evidence scope is narrower than copy, approve narrower wording or keep it unresolved. Record estimates/proxies as such.
  8. Regenerate claims-ledger.md / offers.md only from accepted projection state, then verify claims.

Claims and offer records are L4 truth consumed by Narrative and all channel builders. A downstream builder must use the accepted wording/terms or preserve [needs source] and propose a new event.

Save Results

Require explicit write permission. Ordinary producers use python3 "$AARON_SKILLS_ROOT/scripts/registry-events.py" append claims <proposal.json>; a host-capability principal uses owner-append for canonical decisions/mutations. Never edit the NDJSON stream manually. Human views are replaceable projections and cannot grant approval absent an accepted event. Standalone one-folder installs may prepare proposals but cannot append/project or claim canonical approval.

Reference Materials

Next Best Skill