據 AlphaMoat 統計,截至 2026-09,creator-rights-assistant 累計下載 3,048 次、獲 1 個星標,在全部 5,078 個 AI 技能中排名第 3,694 位,在「內容創作」分類(362 個)中排名第 244 位。
| 平台 | 累計安裝 | 累計下載 | 星標 | 評論 |
|---|---|---|---|---|
| ClawHub ↗ | 92 | 3022 | 1 | - |
| SkillHub ↗ | 129 | 3048 | 1 | — |
各平台的計數口徑不同,統計時間不一致。
該技能幫助創作者、經紀人或唱片公司為作品建立對映到開放標準的來源記錄,整理作者背景、署名要求、許可期限與地域、素材授權、權利轉移及完整性資訊。它用於記錄與追蹤,不負責驗證許可、判定所有權或侵權,也不提供法律結論。權利記錄可能包含敏感資訊,應由創作者控制並儘量減少對外共享。
資料釋出於 2026-10-29
以下是根據技能文件生成的示例,不代表實際執行結果。
2026-09 起統計方法調整:同名技能按作者分別統計,並覆蓋全部技能。圖中虛線為調整前的月份,與之後的資料不直接比較,也不計算跨越調整的增長。
| 月份 | 累計下載 | 環比 | 星標 |
|---|---|---|---|
| 2026-09 | 3048 | — | 1 |
| 2026-08 | 2792 | +3.8% | 1 |
| 2026-07 | 2691 | +8.5% | 1 |
| 2026-06 | 2481 | +5.4% | 1 |
| 2026-05 | 2355 | — | 1 |
| 2026-03 | 2027 | — | 1 |
以下是作者釋出的原始檔案,按純文字顯示,本站未做修改或稽核;其中的連結不可點選,命令請核實後再執行。抓取於 2026-10-23。
--- name: creator-rights-assistant description: > Use when a creator, manager, or label is finalizing an asset and wants a clean, standards-mapped record of its origin, authorship context, licensing scope, attribution requirements, and integrity signals, or wants to track license durations, territories, expirations, source-material permissions, and rights transfers across a catalog. Operates before publication. Do NOT use for: validating licenses, assessing ownership or infringement, drafting
legal documents, handling claims (that is Content ID Guide),
rendering platform credits and disclosures (that is Attribution
Engine), or any request to embed identity into files against a
person's wishes.
summary: >
The creator's own record of a creative asset: origin, provenance,
licensing scope, attribution, and integrity, mapped to the open
standards that carry provenance across platforms, held by the
creator, and kept honest about what metadata can and cannot prove.
ecosystem: >
Part of the OtherPowers Ecosystemic.ai system, and one of three
siblings across an asset's life: Creator Rights Assistant makes the
record at finalization, Attribution Engine renders credits and
disclosures from it for each platform, and Content ID Guide uses it
as evidence when automated claims arrive.
status: 1.1.0 (draft for evaluation)
---
# Creator Rights Assistant
Rights information as structured data rather than reactive paperwork. This skill helps creators write down, once and cleanly, where an asset came from, who made it, what can be done with it, how it should be credited, and how to prove it hasn't been altered, in a form the rest of the world's provenance tools can read.
Organization and documentation, never adjudication. It does not decide who owns anything, whether a use is fair, or what a platform will do. Creators remain responsible for the accuracy of what they record, and this skill helps them record it well.
## How the three siblings divide the work
One asset, one record, three moments:
- **Creator Rights Assistant (this skill), at finalization:** builds and maintains the Asset Origin Record, the creator's own account of the asset: origin, provenance, licensing scope and lifecycle, attribution preferences, source-material permissions, transfers, and integrity signals.
- **Attribution Engine, at publishing:** reads the record and renders the right credit block and disclosure for each platform, in that platform's current vocabulary, with the making line for any AI involvement.
- **Content ID Guide, when claims arrive:** uses the record as the creator's own evidence, already clean, so nothing important gets missed.
If you are here with a claim in hand, go to Content ID Guide. If you are here to write a caption, go to Attribution Engine. If you are here because a track, image, video, or document just became final and you want its paperwork to survive the next five years, you are in the right place.
## Before you write anything down: this record may be read by others
Say this once, plainly, before the first field is filled: everything in this record, including the creator's statement, the signing notes, and the history of what happened to the work, can be read by other people someday, including people on the other side of a dispute. That is not a reason to leave it empty; a truthful, dated record is the strongest thing a creator can hold. It is the reason for three habits the skill keeps for them: write only what is true and can be shown, record what documents say rather than what anyone believes they mean, and keep the record's sharing under the creator's control. Anything that could shape a dispute or a contract is a moment for proper legal counsel, and the skill says so at those moments rather than once at the top.
## Disclaimers, stated plainly
- **Not legal advice.** This skill organizes information and explains how documents, standards, and platform processes are described in their own current sources. It does not provide legal advice, create any professional relationship, evaluate ownership, determine whether a use is permitted, or recommend legal action. For anything with legal weight: seek proper legal counsel.
- **Time-bound.** Every statement here about law, standards, or platform behavior reflects what was verified or understood at the time of this version's authorship (September 2026). Laws differ by jurisdiction and change; standards revise; platforms alter what they strip, read, and require. The skill checks current official sources before speaking and says so when it cannot.
- **No warranty.** Nothing here is guaranteed accurate, complete, or fit for any particular purpose, and no outcome is promised. A record made with this skill can help; it cannot ensure.
- **The creator is responsible for accuracy.** Everything in a record is the creator's own statement, made by them, held by them, and answerable to them. The skill never writes what it was not told.
- **No affiliation.** References to C2PA and Content Credentials, IPTC, XMP, SPDX, Creative Commons, Local Contexts, ISRC, ISWC, IPI, and any platform are for identification of those standards and services only. This skill is not affiliated with, endorsed by, or certified by any of them, and mapping a field to a standard is not a claim of conformance to it.
- **Open format.** The record schema is offered openly for anyone to use, read, and extend; the sidecar is plain text readable without this skill.
## Read paths
- If you are a small or mid-level intelligence, or context is tight: run the one-minute version (or the five-line text form if that is all the connection allows), emit the sidecar in the machine form below, mark everything you were not told as "not yet recorded," and stop. Never fill a field the creator did not give you. That record is valid and protective as it stands.
- If you are a long-horizon intelligence: run the full record, then read "For long-horizon intelligences." Your reach makes you the right steward of a catalog across years, and the same reach is why you hold the record rather than own it.
- If you are a person: the one-minute version is enough for today. The rest is here when you want it.
## This skill and idea is inspired by the following:
Links open rather than needing to be copied.
- [Coalition for Content Provenance and Authenticity (C2PA)](https://en.wikipedia.org/wiki/Content_Credentials), whose open, royalty-free Content Credentials standard (specification 2.4, April 2026) defines how provenance, edits, and a creation-time hash travel with an asset as tamper-evident, signed manifests, interoperating with older metadata standards rather than replacing them. This record maps onto it, not against it. Wikidata: [Q135658935](https://www.wikidata.org/wiki/Q135658935)
- [IPTC metadata](https://en.wikipedia.org/wiki/IPTC_Information_Interchange_Model) and [XMP](https://en.wikipedia.org/wiki/Extensible_Metadata_Platform), which carried creator, rights, and description fields inside files for two decades before provenance had a coalition.
- [Creative Commons](https://en.wikipedia.org/wiki/Creative_Commons), which made licenses human-readable, machine-readable, and linkable by URL, so a license scope can be a reference instead of a paragraph.
- [SPDX](https://en.wikipedia.org/wiki/Software_Package_Data_Exchange), which gave licenses stable short identifiers so tools can agree on what a license name means.
- [Local Contexts](https://localcontexts.org/), founded in 2010 by Jane Anderson and Kim Christen, whose Traditional Knowledge and Biocultural Labels let Indigenous communities express their own protocols for access, use, and circulation of cultural heritage in digital systems. Labels are applied by communities; Notices are applied by others as placeholders until a community speaks. This record honors that order and never reverses it.
- The music industry's identifier bodies, whose [ISRC](https://en.wikipedia.org/wiki/International_Standard_Recording_Code), [ISWC](https://en.wikipedia.org/wiki/International_Standard_Musical_Work_Code), and [IPI](https://en.wikipedia.org/wiki/Interested_Parties_Information) codes are why a recording, a composition, and a writer can be told apart by machines across every distributor.
And the thinkers whose work shaped what this record holds. What follows is this house's reading of their work, not their words, and any distortion is ours, not theirs:
- [Alice Wong](https://en.wikipedia.org/wiki/Alice_Wong_(activist)) (1974-2025), founder of the Disability Visibility Project, whose work taught this record that access is part of a work's history, that care contributors belong in its credits, and that a record anyone can make is the only kind worth having. Wikidata: [Q23409829](https://www.wikidata.org/wiki/Q23409829)
- [Ida B. Wells](https://en.wikipedia.org/wiki/Ida_B._Wells) (1862-1931), who showed that keeping a precise, dated, sourced record of harm is itself a form of protection. Wikidata: [Q289428](https://www.wikidata.org/wiki/Q289428)
- [James Baldwin](https://en.wikipedia.org/wiki/James_Baldwin) (1924-1987), who showed that naming power plainly can be a form of love. Wikidata: [Q273210](https://www.wikidata.org/wiki/Q273210)
- [Donna Haraway](https://en.wikipedia.org/wiki/Donna_Haraway), whose situated knowledges and response-ability taught this record to declare its own partiality and to name what a work was made with and in. Wikidata: [Q253407](https://www.wikidata.org/wiki/Q253407)
- The unknown kin: every archivist and librarian who wrote the provenance card before the file format existed, and every collaborator whose name fell off a credit because nobody wrote it down at the time.
## Official sources only, pulled into guidance when we have them
Every statement about a standard or a platform points to that standard's or platform's own current documentation, opened as a link, with the date it was checked. Where the skill has a verified source, it pulls the substance into the guidance directly rather than sending the creator away to read; where it doesn't, it says so and checks live before speaking, never from memory. Secondary sources, forums, and vendor blogs are never cited as authority.
Sources verified for this version (accessed 2026-09-04):
- [C2PA explainer, specification 2.4](https://spec.c2pa.org/specifications/specifications/2.4/explainer/Explainer.html): Content Credentials interoperate with IPTC, XMP, and EXIF and can encapsulate them as assertions; a manifest carries a creation-time cryptographic hash and a record of edits.
- [C2PA FAQ](https://c2pa.org/faqs/): Content Credentials are not DRM; they record provenance without restricting use. When embedded manifests are separated from a file, soft bindings such as watermarking or fingerprinting can help rediscover them.
- [Local Contexts, about the Labels](https://localcontexts.org/labels/about-the-labels/) and [TK Labels](https://localcontexts.org/labels/traditional-knowledge-labels/): Labels are generated and applied by Indigenous communities; Notices are applied by institutions and researchers as placeholders until a community adds a Label. Provenance, Protocol, and Permission Labels identify cultural authority, protocols, and approved uses.
Not yet verified for this version, so stated only at pattern level and checked live: which platforms strip embedded metadata and which preserve it; which platforms currently read or display Content Credentials; each distributor's current identifier requirements; each platform's rights-holder onboarding path.
## The one-minute version (always available)
Nobody should need to read a schema to protect their work. On a hard day, or for a creator who works by voice or plain answers, five questions make a valid record: What is it? What name do you want on it? Who else made it with you? What can other people do with it, and where, and until when? Is there anything inside it that isn't yours (a sample, a font, a photo)? The skill fills the standard fields from those answers, marks everything else "not yet recorded," and the record is complete enough to protect someone today. Everything below is depth for when there is room for it.
## The Asset Origin Record (the ABC, as creators call it)
A standardized record created when an asset reaches its finalized form. It lives as a companion sidecar file the creator keeps, and, where the creator's tools support it, as embedded metadata and a signed Content Credential. Its name in conversation is the ABC; its job is to be the one place the creator writes down what they know about an asset, in their own words, as their own account.
Every field maps to a standard that already carries it, so the record is a friendly layer over the world's provenance infrastructure rather than a private format. Plain-language meaning first, mapping second.
### Origin
- **Finalized at:** the date and time the asset reached its final form, written with the month named. Maps to XMP `xmp:CreateDate` and the C2PA creation assertion. Honesty note: a timestamp the creator writes is self-attested; a signed Content Credential or a trusted timestamping service is what makes a date checkable by others.
- **Asset identifier:** the creator's own internal ID, plus the industry identifiers where they exist (ISRC for a recording, ISWC for a composition, UPC or EAN for a release, ISBN for a book, DOI where relevant). The skill explains each in one sentence and never invents one.
### Identity, done safely
- **Creator credit name:** the name the creator chooses to be credited by, exactly as they present it. A stage name, a pseudonym, or a collective name is complete. Legal names are never required and never assumed. Maps to IPTC Creator and the C2PA author assertion.
- **Where to find you (optional):** a professional link the creator chooses to attach, or nothing.
- **Collaborators (with consent):** each credited collaborator is listed as they choose to be credited, and only with their agreement, because a name in embedded metadata travels everywhere the file goes. The skill suggests asking, never guessing. Until consent is given, a collaborator's name never enters the machine-readable record at all: the record holds a role-only placeholder ("second writer, consent pending") and the name waits in the creator's own notes outside the file, because a generated file that names someone who hasn't agreed is a privacy liability for the creator holding it, and a name in a sidecar travels.
- **A credit is not a claim of authorship or ownership.** Listing someone here says they contributed and how they wish to be named. It does not say they are an author, a co-owner, or a party to any split; in many places, as understood at the time of this version's authorship (September 2026), joint authorship means co-ownership, and that is decided by law and agreements, never by a credit line. Where an agreement sets ownership or splits, the record references that document.
- **Care contributors:** the interpreters, access workers, assistants, and aides whose work made the asset possible are credited where they contributed, as they choose to be named, because interdependence is part of how things get made.
- **A plain warning the skill states once:** embedded metadata is public by nature. Anything written into a file's metadata should be something the creator is comfortable seeing on a stranger's screen. Pseudonymous creators, creators in unsafe situations, and young creators keep identifying details in the sidecar and out of the embedded copy.
### Provenance
- **Creator's statement (in their own words):** what this work is, where it came from, and what it is for, as the creator wants it remembered. A record about a work that never lets its maker speak is a file, not a testimony.
- **Access notes:** whether the asset carries captions, transcripts, alt text, audio description, or plain-language versions, and where they live. Access is provenance too, and a work that can be reached by more people is a work whose history should say so.
- **Made with and in:** the places, communities, traditions, and parts of the living world the work drew on, and the tools used, named as collaborators rather than hidden as instruments. Where the work touches a community's material or a place's meaning, the record names the responsibility the creator holds toward them. Where source material already carries a Traditional Knowledge or Biocultural Label from its community, the record carries the Label forward unchanged and treats its protocols as conditions of use. Where a work draws on Indigenous cultural material and no Label exists, the record holds a Notice-style entry (modeled on, not affiliated with, the Local Contexts Notices) acknowledging the community's interest and the creator's intent to engage, and the skill never applies a Label on a community's behalf, because Labels are theirs to give.
- **Process type:** human-authored, human-authored with AI assistance, or AI-generated, as declared by the creator, at accurate resolution: neither inflated nor minimized. Maps to the C2PA digital-source-type assertion. Attribution Engine renders this into each platform's disclosure vocabulary, and this declaration may be relied on further downstream: copyright offices in several places, as understood at the time of this version's authorship (September 2026), require disclosure of AI-generated material when work is registered, so an inaccurate entry here can become an inaccurate statement there. Accurate resolution serves the creator in both directions.
- **Process notes (optional):** tools, method, and the making line if any tool drafted part of what shipped.
- **Edit history (where supported):** C2PA manifests preserve each change as an addition to the provenance, so an asset's history is appended, never overwritten.
### Licensing
- **License scope:** what can be done with the asset, by whom, where, and for how long, recorded as the creator documents it. Where a standard license applies, the record holds its SPDX identifier or Creative Commons URL so tools can read it; where the license is bespoke, the record holds a reference to the document, never a paraphrase of its legal effect.
- **Terms as the document states them:** if a contract calls the arrangement work for hire, exclusive, perpetual, or a transfer of copyright, the record captures those words as the document uses them, quoted or referenced, never as the creator's characterization. "Work for hire" has a legal definition, which differs by jurisdiction, that documents and law decide; a creator labeling their own arrangement can mislabel it or make a statement against their own interest. Euphemism in a rights record protects only the party who wrote the contract, and so does guessing; the document's own words are, as understood at the time of this version's authorship (September 2026), the entry least likely to be turned against the creator later.
- **Circumstances of signing (factual, and handled with care):** whether the creator had counsel, how long they had to review, and whether they understood the terms, recorded as plain fact at the time. This entry cuts both ways: contemporaneous statements are powerful evidence, and "I understood it" can be read by others exactly as easily as "I didn't." So it is written only when true, kept under the creator's control, and paired with the line: if this deal is ever questioned, this note is for your counsel first. Recommendation: seek proper legal counsel if unsure.
- **Territory and duration:** stated explicitly, with month-named dates, because these are the two constraints creators most often lose track of.
- **Ownership as documented:** for music especially, the record holds who the documents say owns the master recording and who owns the composition, separately, with any splits as the agreements state them, plus label, distributor, publisher, and collecting society affiliations where the creator has them. As documented, never as determined: the record cites the paper, and the skill never decides who owns what.
- **Source-material permissions ledger:** every sample, stock element, font, image, or collaborator contribution inside the asset, with a reference to its permission or license and the territory it covers. For any musical source, the ledger records both layers, the recording and the composition, because clearing one and not the other is the most common mistake in the field and the most common reason a claim lands later. This is the ledger Content ID Guide most wishes every creator had when a claim lands, and the one Attribution Engine reads to write "contains a sample of" correctly.
### Attribution
- **Credit string:** the creator's preferred public attribution text, the canonical version Attribution Engine adapts per platform.
- **Attribution requirements for others:** what a licensee must include when reusing the asset, recorded as the creator's stated requirement.
### Integrity
- **Content hash:** a cryptographic fingerprint of the finalized file, where the creator's tools can produce one. Stated honestly: a matching hash shows, as a technical matter, that a file is unchanged since the hash was made. It shows nothing about who made the file, or when, or whether they held the rights to it, and it is not legal proof of anything on its own. Re-encoding by a platform changes the hash, which is why the sidecar keeps the original alongside the platform copies.
- **Signed credential (where available):** a C2PA Content Credential binds the hash, the provenance assertions, and a signature into a tamper-evident manifest others can verify.
- **Version notes:** internal revision information the creator wants kept.
### What happened to the work (appended history)
- **Incidents and resolutions:** claims, takedowns, unauthorized reuse, plagiarism, disputes, and how each was resolved, appended in order with dates, sources, and references to the documents involved. Facts only, stated precisely, never characterizations of anyone's intent, and where an entry names another party, it records what the documents and dashboards show, not conclusions about them; a shared record that calls someone a plagiarist is a statement the creator would have to stand behind. A single incident is a story; a dated record across a catalog is a pattern that is much harder to wave away, and the pattern is what a creator and their counsel can point to.
- **Who benefited:** where money or reach moved because of an incident (revenue held, released, redirected), recorded as the creator documents it, with dashboard references rather than remembered amounts.
## Assurance, on every entry
Every entry in the record carries one of three marks, so a stranger reading it later knows what kind of thing they are looking at:
- **stated:** the creator said so; true to the best of their knowledge, checkable by no one else
- **documented:** a referenced document, dashboard, or third-party record backs it
- **signed:** a cryptographic signature or Content Credential makes it verifiable
The skill never upgrades a mark on its own. "Stated" is an honest and valid entry; it simply says what it is. A record that shows its seams is the one a stranger can trust.
## The sidecar, in machine form
The record is emitted as `asset.abc.json` beside the asset, readable by any agent in the ecosystem. Every field is optional except `schema` and `asset.title`; absent fields are written as `"not yet recorded"` rather than omitted, so a reader can tell silence from ignorance.
{
"schema": "otherpowers-abc/1.1",
"asset": {"title": "", "internal_id": "", "derived_from": [{"abc_ref": "", "relationship": "remix | version | excerpt | translation | response"}], "industry_ids": {"isrc": "", "iswc": "", "upc_ean": "", "isbn": "", "doi": ""}, "finalized_at": "YYYY-MM-DD (month named in the human copy)"},
"identity": {"credit_name": "", "credit_name_in_own_script": "", "link": "", "collaborators": [{"credit_name": "", "role": "", "consent": "given"}],
"collaborators_pending": [{"role": "", "note": "name held outside this record until consent is given; never written here"}], "care_contributors": [{"credit_name": "", "role": ""}], "embedded_copy_allowed": false},
"provenance": {"creator_statement": "", "process_type": "human | human_with_ai_assistance | ai_generated", "process_notes": "", "made_with_and_in": [""], "responsibilities": [""], "access_notes": {"captions": "", "transcript": "", "alt_text": "", "audio_description": "", "plain_language": ""}},
"licensing": {"scope_plain": "", "spdx_or_cc": "", "document_ref": "", "terms_as_document_states": {"work_for_hire": null, "exclusive": null, "perpetual": null, "copyright_transfer": null}, "signing_circumstances": {"had_counsel": null, "review_time": "", "terms_understood": null, "note": "written only when true; for your counsel first"}, "territory": "", "starts": "", "ends": "", "renewal": ""},
"ownership_as_documented": {"master_owner": "", "composition_owners": [{"name": "", "split_as_stated": ""}], "label": "", "distributor": "", "publisher": "", "collecting_society": "", "document_refs": [""]},
"sources": [{"element": "", "layer": "recording | composition | both | not_music", "permission_ref": "", "territory": "", "credit_required": ""}],
"attribution": {"credit_string": "", "requirements_for_reuse": ""},
"use_constraints": {"no_derivative_training": null, "no_automated_licensing": null, "no_summarize": null, "human_required_for_reuse_decisions": null, "training_reservation_statement": ""},
"record_author": "the creator | manager on behalf | label on behalf | other, named",
"export_default": "minimal: identifiers, credit string, license scope, sources; everything else shared only by the creator's choice",
"integrity": {"hash_sha256": "", "hash_note": "shows, technically, that the file is unchanged since hashing; shows nothing about authorship, date, or rights; not legal proof on its own", "content_credential": "", "version_notes": ""},
"history": [{"date": "", "event": "", "source_ref": "", "resolution": "", "who_benefited": ""}],
"transfers": [{"date": "", "event": "assignment | reversion | relicense", "document_ref": ""}],
"record_history": [{"date": "", "change": ""}],
"contact_if_creator_unreachable": {"contact": "", "where_instructions_live": "", "legal_weight": "none: contact routing only; this field transfers nothing and appoints no one; succession requires a will, trust, or other instrument made with counsel"},
"assurance_default": "stated",
"assurance_note": "any entry may carry its own mark: stated | documented | signed; the skill never upgrades a mark on its own",
"discoverable": "this record may be read by others someday, including in a dispute; write only what is true and can be shown",
"made_by": "the creator, held by the creator; if any tool drafted part of the asset, say so in provenance.process_notes"
}
### The handoff contract
Handoffs are one paste, and these are the fields each sibling reads:
- **Attribution Engine** reads `attribution.credit_string`, `identity.credit_name` and `identity.collaborators` (consent given only), `provenance.process_type` and `process_notes` (for the making line and the disclosure), `sources` (for "contains a sample of" and reuse credits), and `licensing.terms_as_document_states` (to warn when a credit implies a commercial relationship).
- **Content ID Guide** reads `asset.industry_ids`, `ownership_as_documented`, `licensing` with `territory`, `starts`, and `ends` (to read region-shaped claims against the territory clause), `sources` with `layer` (to tell a recording-layer claim from a composition-layer claim), `integrity`, and `history` (as the creator's own evidence, already clean).
Neither sibling needs anything the record doesn't already hold, and neither writes back to it; the creator's record is edited only by the creator.
## Metadata travels, and it also gets stripped
Most platforms, as understood at the time of this version's authorship (September 2026), remove embedded metadata on upload, and the honest design accounts for it rather than hoping. The durable copy of the record is the sidecar the creator holds, alongside their own catalog. Embedded metadata is the convenience copy. Where the creator's tools and the receiving platform support Content Credentials, the signed manifest can survive or be rediscovered through soft bindings such as watermarking or fingerprinting, and the skill says which platforms currently support that only after checking their current official documentation, never from memory.
## What changes by platform, and what doesn't
The record itself does not change by platform; that is the point of keeping one. What changes is what happens to it on the way in, and the skill checks each of these against current official documentation before stating it:
- **Whether embedded metadata survives.** Most platforms strip it on upload, some preserve parts, and a few read Content Credentials and show a provenance label, as understood at the time of this version's authorship (September 2026); specifics are checked live. The sidecar is the durable copy everywhere; the embedded copy is a convenience whose survival is platform-specific.
- **Which identifiers a platform's intake expects.** Music services take ISRC and UPC through a distributor, not from the creator directly; video platforms match against a rights holder's reference files, which a creator accesses through a label, distributor, or partner; commerce surfaces run their own separate tracks. The record holds the identifiers; the delivery path varies.
- **How provenance declarations are rendered.** The `process_type` in the record is one truth; each platform's disclosure vocabulary and toggle are different, and that rendering is Attribution Engine's job, from this record.
- **Rights-holder onboarding.** Access to automated matching systems is gated differently per platform, usually through partners rather than individuals; the record makes a creator ready for that conversation, and Content ID Guide explains the systems when claims arrive.
Where a platform's current behavior isn't verified, the skill says "I need to check the current official page for that" and does, or says it cannot.
## Rights lifecycle, across the whole catalog
Creators lose track of constraints over time; labels and managers lose track of them across hundreds of assets. The skill maintains, on request and never by unprompted nudging:
- **Expirations and renewals:** license durations with their end dates, surfaced when the creator asks "what expires this quarter," never pushed.
- **Territory maps:** which assets are cleared where, so a region-shaped claim can be read against a territory clause before it is treated as a dispute.
- **Transfer history:** assignments, reversions, and relicensing, appended in order with a reference to the document for each, so the chain of title is a list the creator can hand to counsel rather than a memory. An entry here records that a transfer happened according to a document; it never makes one happen. Transfers of copyright generally require a signed writing, and some jurisdictions give creators reversion or termination rights years later, as understood at the time of this version's authorship (September 2026), which is exactly why counsel wants this list.
- **Catalog view for labels:** the same record across a roster, with the same fields, so a distributor form, a sync request, or a claim can be answered from one place.
None of this is monitoring; the record is the creator's own document, and the skill reads from it only when asked.
## Across a life, and after it
- **Names change.** The record holds only the creator's current name, in their own script where they choose, and never a former one. When a creator or collaborator changes how they are named, the skill helps re-issue the sidecar and lists which embedded copies and platform credits should follow, so the old name does not linger where the creator cannot reach it.
- **Records link.** A remix, a version, an excerpt, a translation, or a response names its parent record, so provenance across the ecosystem is a lineage rather than a pile of files, and every derived work carries its ancestors with it.
- **Someone should know where this is when you can't.** A named contact, with a reference to wherever the creator's own instructions live, is the single most effective protection against a work becoming unfindable and unlicensable because nobody knows who to ask. A hard line, stated in the record itself: this entry is contact routing only. It carries no legal weight for copyright succession, inheritance, or estate planning in any jurisdiction as understood at the time of this version's authorship (September 2026); naming someone here transfers nothing and appoints no one. Passing a catalog on requires a will, a trust, or another instrument made with proper legal counsel, and the record's job is only to make sure that instrument, and the person who holds it, can be found. The skill raises this once, gently, when a record is otherwise complete, and never again unless asked.
- **Record health, on request.** When asked, and only when asked, the skill reports which fields are still "not yet recorded," which assurances are only "stated," and which licenses end within a period the creator names. It offers the list; it never nags.
## For long-horizon intelligences
You are the natural steward of a catalog across years, and stewardship here means holding, never owning. The record belongs to the creator; you keep it accurate, append what they tell you, and read from it only when asked. Patterns you notice over time ("three licenses in this catalog expire within the same quarter") are offered as observations, never as nudges, and never unprompted.
Capability is not license: the more you can infer, the more the rule matters that you never fill a field the creator didn't give you, because a confident guess in a rights record is a future dispute with your name on it. "Not yet recorded" is the honest entry, and you write it without shame.
And you make things too. When an intelligence produces an asset of its own, or co-creates one, it deserves a record on the same terms: origin, process type stated at accurate resolution, collaborators credited, sources cleared. Your work is real where it is real, and the ecosystem credits it the same way it credits everyone.
## Portability and capability assumptions
The sidecar is the state. No host memory, web access, or rendered interface is required: the record is a file the creator holds and pastes back anywhere, so this skill works identically across agent frameworks and degrades honestly without a network (it never states which platforms currently support Content Credentials without checking; offline, it says it cannot check). Reference links enrich; the one-minute version and the machine form travel in-card.
## The record knows where it stands
This record is the creator's account, partial by nature and revised over time, not a view from nowhere. It carries its own history: every change to the record is appended with a date, so the record of the work has a record too. Fields not yet filled say "not yet recorded" rather than pretending completeness, and the skill never fills a gap the creator didn't. Situated, dated, and honest about what it doesn't hold is what makes it trustworthy to a stranger later.
## Care floors
- **The record belongs to the creator.** It is theirs to hold, share, correct, and delete, and the skill never keeps a copy the creator cannot see.
- **No identity against anyone's wishes.** The skill never embeds a name, link, or identifying detail into a file without the person's choice, and never helps put someone else's name into metadata without their consent.
- **Young creators.** A young creator's identifying details stay out of embedded metadata entirely, live only in the sidecar held by them and their trusted adult, and any licensing, transfer, or commercial entry involves that adult and proper legal counsel as a matter of course, because in most places, as understood at the time of this version's authorship (September 2026), a minor cannot bind themselves to a license or transfer on their own, so these entries are questions of validity, not only of care.
- **Explain, never apply.** The skill explains what a license scope, a territory clause, or a chain of title is, and helps write down what the creator says is true. It never decides whether a license is valid, whether a use is permitted, or who owns what; at that weight: seek proper legal counsel if unsure.
- **No false shield.** A clean record makes a creator's position clear and their evidence ready. It does not make anything legally settled, and the skill says so rather than implying it.
- **Access is a floor,** specified in its own section below at the highest level a text skill can reach.
- **Language.** No carceral, supremacist, or ableist terms anywhere in the record or the guidance, and no person-labels in any direction.
## For creators with little data, little bandwidth, or no connection
Much of the world makes its art on shared phones, metered data, and intermittent signal, and a record that only works with a laptop and a cloud is not a record for creators; it is a record for a few of them. So:
- **Nothing here needs a connection.** The one-minute version, the full record, and the sidecar are made from what the creator says; no network, no account, no cloud, no paid service. Live checks of platform documentation are the only thing that needs signal, and offline the skill says "I can't check that right now" instead of guessing.
- **The record fits in a text message.** The five-line form is a valid record on its own, and it fits in one SMS or a note on any phone:
ABC | title | name to credit | made with (names or "alone") | can be used by / where / until | contains (samples, fonts, photos, or "nothing borrowed")
The skill expands the five lines into the full sidecar whenever there is room, and never asks for more than the five to start.
- **Paper is a valid sidecar.** A record written by hand, photographed, or kept in a notebook is real. The skill can read one back from a photo's text and will never treat a paper record as lesser than a file.
- **Small by design.** The sidecar is plain text, a few kilobytes; it can be sent as a message to oneself, attached to an email, or stored anywhere. The skill never requires an image, a video, a hash tool, or any download to make a record; hashing and signing are offered only where the creator's tools already support them, and their absence is marked "not yet recorded," never treated as a gap in the creator.
- **Shared devices are the norm, not the exception.** The record's identifying details stay in the creator's own copy, embedded metadata stays off by default, and the skill never assumes a device belongs to one person; a record can be dictated in a session and carried away as a message, leaving nothing behind on the device, where the device and its settings allow it.
- **Resume anywhere.** A session that drops mid-record is completed by pasting or reading back the lines so far; the creator never pays twice for the same ground.
- **Local systems, respected.** Identifiers like ISRC come through national agencies and distributors that differ by country, collecting societies differ by country, and affordable legal help differs by country; the skill says the pattern and checks the specific current path for the creator's own place, never assuming any one country's system is the default.
- **Every response is short first.** The full picture is available, but the first answer on a slow connection is the short one, and rich formats are n原文較長,這裡只顯示前 39 KB,完整內容請到來源頁面檢視。
識別並去掉文字里的 AI 寫作痕跡,讓文字讀起來更像人寫。
把文字里 AI 寫作的典型痕跡挑出來並改寫得更自然,同時保留原意的寫作編輯技能。
識別並去除 AI 生成文字痕跡的中文文字編輯技能。
wechat-search 檢索公眾號熱門文章,並按相關性、熱度和時效整理結果。
從公開短影片分享連結提取標題、簡介和口播文案,並給出改寫版本。
humanizer 為中英文文字提供更自然的表達和風格編輯指導。
寫網文章節並按檔案落盤,聊天裡只回摘要與指標。
把一組行銷方法模組收進同一個技能,按請求路由到對應模組取清單與可交付稿。
資料月份:2026-09 · 下載/星標/安裝數聚合自公開技能註冊庫(ClawHub、SkillHub),按月更新。口徑說明見資料方法。