Digital Asset Management

Digital Asset Management

Use this skill for the media-specific lifecycle and vocabulary that connect a photo/video catalog, its viewers, its editing workflow, and its storage. Inspect the product's supported formats, storage model, existing media libraries, and privacy policy before choosing representations or changing behavior.

Scope And Vocabulary

• Asset: the logical, user-visible media item. Do not equate it with a path, filename, byte sequence, or rendered file.
• Media file: stored bytes associated with an asset, with a role such as original, imported copy, sidecar, derivative, proxy, preview, or export.
• Original: immutable source bytes and their import provenance. Preserve them unless the product explicitly defines replacement semantics.
• Rendition: generated output for a defined purpose, such as a thumbnail, preview, video proxy, or export.
• Edit recipe: versioned, ordered operation parameters that can reproduce a rendition from a source. It is not the same as an undo history or an asset version.
• Asset version: a user-visible revision or branch that can be restored. Keep it distinct from metadata revisions, edit-recipe revisions, and renderer versions.
• Folder: a physical/import location or hierarchy. Manual collection: a user-managed membership list. Smart collection: a saved, validated query. Tag/keyword: a reusable classification. Do not collapse these concepts.

Workflow

1. Identify supported image/video formats, storage topology, media-processing libraries, source-of-truth locations, and background-worker boundaries.
2. State lifecycle rules before editing: ingest, duplicate detection, metadata extraction/write-back, editing, rendition generation, sharing, deletion, retention, and restore.
3. Define domain nouns and invariants before choosing tables, endpoints, or UI controls. Model an original, an edit recipe, a rendition, and a user-visible version independently when the product needs each one.
4. Define authority and conflict behavior for embedded metadata, sidecars, catalog edits, external file changes, concurrent edits, and ACL changes.
5. Define viewer and organization behavior for loading, unsupported media, orientation, color, keyboard interaction, hierarchy moves, and selection.
6. Route implementation mechanics to the owning skills, then add focused tests for the invariants and user-visible workflow that changed.

Ingest And Identity

• Preserve immutable original bytes, source location/provenance, extraction result, and import failure separately from generated output.
• Treat byte hash, perceptual similarity, filename, and logical asset identity as different facts. Define duplicate, re-import, moved-file, and replacement behavior explicitly and make ingestion idempotent where retries are possible.
• Treat uploaded/imported media as untrusted input. Determine format from byte signatures and parser-confirmed structure against an allowlist; do not trust extensions or caller-supplied MIME types, and reject material mismatches.
• Bound compressed and expanded bytes, dimensions/pixels, duration, streams, frame count, metadata depth/count/value size, CPU/wall time, memory, temporary and output storage, and queueing per item and per tenant or workload boundary.
• Process risky formats and codecs with least privilege and operating-system, container, or VM isolation proportionate to the threat model. Do not expose ambient credentials or unnecessary filesystem/network access. Invoke reviewed external tools with structured argument APIs, never through a shell command assembled from media paths or metadata.
• Model asynchronous extraction/rendition work with observable states, cancellation, retry, and terminal failure behavior.

Metadata And Interpretation

• Preserve raw embedded and sidecar metadata where interoperability requires it; extract normalized values for search/display without losing provenance.
• Define field-level precedence and conflict handling among EXIF, IPTC, XMP, ICC, sidecars, catalog edits, and later external changes. Do not overwrite a whole metadata block merely to change one field.
• Apply orientation consistently to previews, viewer dimensions, crop coordinates, exports, and API values. Distinguish stored from display dimensions.
• Make timestamp timezone assumptions, repeatable values, legacy encodings, GPS precision, and missing values explicit.
• For video, distinguish the container from video, audio, subtitle, and data streams. Record timebase, display rotation, duration, codecs, HDR/color data, and keyframe limits only when the product needs them.

Read metadata and interoperability when this boundary is material.

Non-Destructive Editing And Renditions

• Store validated operation parameters and their order as an edit recipe; retain original bytes. A preview, proxy, or export is generated output, not authority for future edits.
• Define a canonical coordinate space for crop/geometry after orientation policy. State units, ranges, defaults, and rounding.
• For automatic operations, persist resolved parameters and algorithm/model version. An unversioned auto flag cannot reproduce a result.
• Include source identity, recipe revision, presentation-affecting metadata, color/output profile, and renderer/codec version in rendition-cache decisions. Invalidate atomically and do not serve stale or unauthorized output.
• Define whether undo/redo, restoring an asset version, and branching edits move pointers or create new immutable revisions.

Read editing and rendering for operation, cache, export, and video-specific details.

Organization, Hierarchies, And Viewing

• Give manual collections, folders, tags, and smart collections separate semantics. Specify ordering, membership, root, delete, and authorization rules.
• Give smart collections a versioned, validated query language. Define supported fields/operators, null versus absent data, timezone/collation/full-text rules, live versus snapshot membership, deterministic sorting, and schema evolution.
• Prevent hierarchy cycles, including concurrent self/descendant moves. Define sibling names, reparenting, subtree deletion, and parent authorization.
• Use thumbnails/proxies appropriate to the view; do not decode originals for catalog browsing by default. Define progressive loading, cancellation, error, unsupported-media, and stale-preview states.
• Verify orientation and color correctness, virtualized grid/filmstrip behavior, zoom/pan, accessible selection, keyboard tree navigation, and video playback controls/captions where applicable.

Read catalog, persistence, and organization and viewer, organization, and testing when designing these areas.

Sharing And Read Paths

• Enforce tenant and object authorization on originals, metadata, renditions, exports, search/index results, background jobs, and cache hits. Client-visible identifiers, collection membership, or possession of an old URL are not authorization.
• Treat signed/share URLs as bearer capabilities. Bind them to the tenant, asset/version/rendition, purpose, method, and a short lifetime; define revocation and do not put capability values in logs, analytics, referrers, or durable client-visible locations beyond the intended handoff.
• Propagate ACL, privacy, version, and deletion transitions to every normal read path and generated representation. Expire capabilities and purge or deny stale caches/exports; keep retained backups inaccessible to normal reads and remove them under the documented retention schedule.
• Serve only allowlisted response media types. Set X-Content-Type-Options: nosniff, an appropriate Content-Disposition, and cache directives matching sensitivity and revocation needs; do not render attacker-controlled active content inline merely because it was uploaded as media.
• Test access before and after tenant, ACL, privacy, version, share, and deletion transitions, including direct IDs, old capabilities, caches, indexes, workers, thumbnails, previews, proxies, and exports.

Skill Routing

• Use domain-driven-design for bounded contexts/invariants and domain-modeling for an implemented model review.
• Use sql-engineering, then postgresql-sql-engineering, mysql-mariadb-sql-engineering, or sqlite-sql-engineering, for schema, migration, query, and transaction mechanics.
• Use api-design for ingest, metadata, edit, collection, rendition, pagination, and concurrency contracts.
• Use css-scss-styling for viewer styling, ux-accessibility-review for rendered viewer/tree review, and playwright-e2e for browser-visible workflows.
• Use performance-review for catalog-scale browsing, rendering, caching, and background-workload evidence.
• Use security-review and threat-modeling for parsing, uploads, sharing, metadata exposure, derivatives, URLs, caches, and retention. GPS, face/person tags, camera identifiers, timestamps, captions, and audio may be sensitive.
• Use test-driven-development and testing-strategy for test design and levels.

Completion Checklist

• Original, asset, rendition, recipe, and asset-version semantics are distinct.
• Ingest, duplicate, metadata-conflict, deletion, and restore behavior is stated.
• Hierarchy cycles and smart-collection query semantics are prevented or defined.
• Previews, proxies, originals, and exports apply current authorization and privacy rules independently.
• Tests cover orientation, metadata precedence, hierarchy invariants, idempotent ingest, edit/version distinction, and rendition invalidation where relevant.
• Performance evidence uses representative catalog sizes, media fixtures, and bounded decode/render workloads.

Non-Goals

Do not prescribe a codec, image-processing library, cloud provider, ORM, schema, or UI framework. Do not use this skill as a replacement for generic SQL, API, language, security, or styling guidance, and do not apply it to ordinary document management without photo/video asset lifecycle requirements.