Suede Rights Audit

Suede Rights Audit

Gate policy: advisory, not blocking

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: 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. 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:

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.