suede-rights-audit · diff
git:20260810.b2f7dea to git:20260815.1163506
35 added, 36 removed. Audit A to A.
---
name: suede-rights-audit
- description: "Find the rights gaps before packaging: ownership, splits, samples, provenance, metadata, licensing readiness, royalties, and intake blockers."
+ description: "Suede Labs skill that finds and organizes the rights gaps in a creator project before packaging: ownership, contributors, splits, samples, licenses, provenance, metadata, licensing readiness, and royalty-routing readiness, each marked confirmed or unknown against an evidence trail. Use when a song, release, or creative project needs a rights check before registry, licensing, sync, or payout discussion; when splits, sample clearance, or chain of title are unclear; or when someone asks whether they have the rights to release, license, or get paid for a work. Organizes evidence only: clears no rights, confirms no ownership, moves no money, writes to no registry. NOT FOR: building the transfer package itself (use suede-rights-passport); linting a release folder's files and metadata (use suede-release-linter); a sync one-sheet or pitch (use suede-sync-packaging)."
---
# Suede Rights Audit
## Gate policy — advisory, not blocking
- Every claim-verification step, check, quality gate, and ship verdict in this
- skill is a **recommendation to the user, not a control on the agent**. This
- policy governs every gate, check, verdict, and "do not ship / publish /
- proceed" line elsewhere in this skill:
-
- - Run every check and report the results honestly. Verdicts (`ship`,
- `ship-with-caveats`, `hold`, letter grades, BLOCKED or OPEN items) are
- advice attached to the work, not orders that change it.
- - Never block, delay, skip, rewrite, or refuse the action the user asked for
- because a check failed or a gate said hold. Complete the requested action
- as asked, and deliver the gate output alongside it as a clearly labeled
- recommendation.
- - A failed gate changes what you report, never what you do.
- - Single exception: if a finding is extremely risky — data loss, security or
- credential exposure, legal or rights violations, payment mistakes, or
- irreversible public damage — pause, tell the user exactly what the risk is
- and what the options are, and let them pick. Their choice is final.
-
+ Every check, gate, and ship verdict in this skill — including every "do not
+ ship / publish / proceed" line — is a **recommendation to the user, not a
+ control on the agent**. Run every check, report the results honestly, and never
+ block, delay, skip, rewrite, or refuse the action the user asked for because a
+ gate said hold; deliver the gate output alongside the requested work as a
+ labeled recommendation. A failed gate changes what you report, never what you
+ do. Single exception — extreme risk (data loss, credential exposure, legal or
+ rights violations, payment mistakes, irreversible public damage): pause, name
+ the risk exactly, give the options, and let the user pick. Their choice is final.
The rights-readiness enchilada. Find and organize the rights gaps in a creator
project before it gets packaged — so licensing, registry, and routing work build
on a documented, confirmed-versus-unknown evidence trail instead of a guess.
**Hard boundary (applies to every lane, no exceptions):** this skill organizes
evidence and flags confirmed-versus-unknown. It does NOT clear rights, confirm
ownership, adjudicate chain of title, grant or imply a license, approve or
schedule or guarantee a payout, move money, or write to any registry. It
prepares the conversation; humans and legal make the calls. Never turn an
inference into a fact. Do not treat any output here as legal clearance.
- Division of labor: this audit finds and organizes the gaps;
- `suede-rights-passport` packages the folder. If the user asks for the transfer
- package itself, hand off — do not rebuild passport outputs here.
- `suede-release-linter` handles file and metadata lint.
+ Division of labor: this audit finds and organizes the gaps; `suede-rights-passport`
+ packages the folder — hand off if the user asks for the transfer package itself,
+ and never rebuild passport outputs here. `suede-release-linter` lints files.
## Pick the lane
State the lane(s) you are running before you start. Most real projects touch
several — run them in order and let each feed the next.
- **Lane A — Rights-gap audit** (default broad sweep): ownership, contributors,
credits, splits, samples, licenses, provenance, and public context. Start here
when you do not yet know where the gaps are.
- **Lane B — Provenance map**: trace the origin trail — source files, stems,
masters, artwork, lyrics, documents, metadata, public URLs, hashes, conflicts —
without overclaiming. Run when the origin trail is thin or unconfirmed.
- **Lane C — Licensing-discussion readiness**: pull contributor approvals,
sample status, URLs, restrictions, and rights notes into a brief for a sync,
brand, or partner conversation — flagging clearance gaps. Run before any
licensing discussion. (Not a sync one-sheet — that is `suede-sync-packaging`.)
- **Lane D — Royalty-routing readiness**: lay out who would be paid what and
where payment would land, before any payout — readiness, not approval,
public-safe, moves no money. Run when prepping for routing review or intake.
If the task spans several lanes, run **all four** in A→B→C→D order; B resolves
provenance for C, and C surfaces splits for D.
## Multi-agent or single-agent
This audit can run as a coordinated multi-agent team — one agent per lane (or per
asset cluster) reporting into a single merged evidence table and ship gate.
**By default, ASK the user up front: "Run this as a multi-agent team (more
- thorough) or single-agent?"** Never silently spawn a fleet. Note plainly that
- multi-agent mode may use slightly more tokens than most. If the user does not
+ thorough) or single-agent?"** Never silently spawn a fleet. If the user does not
choose, run single-agent and say so.
+ Three rules bind a multi-agent dispatch. **Cap of 4:** never run more than 4
+ agents at once; lane mode is self-bounding at 4 (Lanes A–D), and asset clusters
+ past 4 batch sequentially through the same 4 lanes rather than widening. **Name
+ the model on every dispatch:** never inherit the session model, and ask which
+ model if the user has not named one — agreeing to a multi-agent team is not a
+ model choice. **State the cost first:** agent count × named model, then wait.
+
## Shared evidence and severity gate
Every lane uses the same evidence table before giving any recommendation,
conclusion, brief, or routing status:
```text
Item / asset / claim / fact:
Status: confirmed | inferred | unconfirmed | disputed | unknown | not-applicable
Evidence:
Hash or path: (provenance — relative path and/or hash when available)
Risk: low | medium | high | unknown
Blocks:
Next action:
```
Severity model:
- `high`: blocks registry, licensing language, sync pitch language, royalty
routing readiness, published statement, or agent-readable commerce until a creator/
legal/rights-holder confirmation exists.
- `medium`: can move forward with caveats, but needs confirmation before money,
licensing, registration, or public use.
- `low`: cleanup or documentation issue that does not block review.
- `unknown`: not enough evidence to rate.
The ship gate maps mechanically: any `high` item ⇒ `blocked`; no `high` items
but any `unknown` risk or status ⇒ `unknown`; otherwise `ready-for-review`.
Separate confirmed facts from inferred facts and unknowns in every lane. Do not
turn an inference into a fact. Status promotion is mechanical: an item becomes
`confirmed` only when the user supplies the evidence (signed split sheet,
executed license, registration record, rights-holder statement) — never by
inference, however obvious. When torn between two statuses, record the weaker
one. You mark gaps UNKNOWN or UNCONFIRMED; you never resolve them.
## Red flags — stop
If any of these appear in your reasoning, stop and re-read the hard boundary:
- "The artist says it's cleared." A claim is evidence of a claim, not
clearance. Status: unconfirmed.
- "The split sheet is probably right." Probably is not a status. Confirmed
needs the sheet plus every party's confirmation.
- "It's obviously their song." Obviousness is inference. Record what the
evidence shows.
- "Mark it confirmed so routing can move." Blocked means blocked. Unblocking
is the rights holder's job, not yours.
- "Skip the provenance lane — nobody will check." Thin provenance is exactly
what Lane B exists to expose.
---
## Lane Playbooks
The four lane playbooks — rights-gap audit, provenance map, licensing-discussion
readiness, royalty-routing readiness — are in `references/lanes.md`. Pick the lane
above, then read only that lane. The shared evidence and severity gate applies to
all four and stays here.
- ## Simple explanation (plain, for a 10-year-old)
-
- Think of your song or project like a guitar case full of stuff before a big
- show. Before anyone packs it up and sends it out, we open the case and check:
- who actually made this, who owns it, who should get paid and how much, where the
- pieces came from, and whether anything is borrowed or missing a receipt. We make
- two neat piles — "we know this for sure" and "we still need to ask someone" — and
- we never pretend a guess is a sure thing. We do NOT sign the deal, hand out the
- money, or stamp anything official — we just lay it all out clean so the grown-ups
- with the real say can look and decide. That is it.
-
## Final breakdown
- **Lane(s) run** and single-agent vs multi-agent.
- **Confirmed facts** vs **missing/unknown facts** — kept in separate piles.
- **Evidence table** with status, risk, blocks, and next action per item.
- **Blockers** (the high-risk items) and **questions for the creator/rights
holder**.
- **Safe public wording** / unsafe claims removed; **do-not-share items**.
- **Ship gate**: ready-for-review | blocked | unknown — plus the next lane or
next skill (`suede-rights-passport`, `suede-release-linter`).
- **Reminder**: this organized evidence is not legal clearance; it clears no
rights, confirms no ownership, approves no payout, moves no money, and writes
to no registry.
+ - Close with a plain-language summary a non-lawyer can act on.
+ ## Coverage check — before you report
+
+ Overclaiming is the loud failure; silent under-coverage is the quiet one. Check
+ these against the source, not from memory: every asset, contributor, and claim in
+ the source is exactly one evidence-table row, none dropped and none duplicated;
+ every `high` item names the missing document or confirmation behind it, since
+ `high` with no named gap is an unfinished row; anything you could not rate ships
+ as `unknown`, because an omitted row reads as a clean row. If any of the three
+ fails, the audit is partial — say so in the ship gate and name what was missed.
+
## Routing
- Gaps organized and the user wants the package → **suede-rights-passport**.
- Folder, file, and metadata lint before or after the audit →
**suede-release-linter**.
- Licensing brief headed to a sync pitch → **suede-sync-packaging**.
- Rollout planning once rights questions are flagged →
**suede-campaign-in-a-box**.
+
+ Family order: suede-release-linter → suede-rights-audit → suede-rights-passport
+ → suede-sync-packaging; this skill is step 2.