agentsclimarketplace

Media

Skill markusleben/ha-nova/skills/media

Use when controlling or browsing Home Assistant media players — play, pause, volume, source/sound mode, speaker grouping, browsing media libraries, and TTS announcements — through HA NOVA Relay.From its SKILL.md

Install
npx -y skills add markusleben/ha-nova --skill media

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 21 stars21 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its file declares

Copied from the file, not written here

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

10.4 KB, as published. Nobody here has run it

HA NOVA Media

Scope

Media player control and discovery:

  • transport control (play, pause, stop, next, previous, seek), volume, mute, source and sound mode
  • browsing media libraries and media sources, then playing a resolved item
  • speaker grouping (join / unjoin)
  • TTS announcements to a player

Not in scope: creating automations that use media players (ha-nova:write), camera streams (ha-nova:camera), generic service calls to other domains (ha-nova:service-call).

Bootstrap (once per session)

Read and follow ../ha-nova/session-bootstrap.md. Verify relay CLI: ha-nova relay health If this fails: ha-nova setup

Relay Contract

  • ha-nova relay ws --data-file <payload-file> for WS commands (browse, search)
  • ha-nova relay core --method POST --path /api/services/<domain>/<service> --body-file <payload-file> for service calls
  • ha-nova relay core --method GET --path /api/states/<entity_id> to read a player's state and attributes
  • --out <result-file> for large browse responses; --jq-file <filter-file> for non-trivial filters

Feature Gate (read before acting)

A media player's attributes.supported_features bitmask decides what it can do — never assume. Read the entity state first and check the bit before offering an action:

BitFeature
1pause
2seek
4volume set
8volume mute
16previous track
32next track
128turn on
256turn off
512play media
1024volume step
2048select source
4096stop
8192clear playlist
16384play
32768shuffle set
65536select sound mode
131072browse media
262144repeat set
524288grouping
1048576announce
2097152enqueue
4194304search media

If a requested feature bit is absent, say so plainly instead of calling the service and reporting a silent no-op.

Flow

  1. Resolve the player yourself — this skill owns its resolution (one skill per intent; never dispatch to another skill mid-task):
    • exact media_player.<id> from the user: use it directly.
    • name or room: WS {"type":"config/entity_registry/list_for_display"}, filter ei starting with media_player., and match on ei/en. For room phrasing, resolve the area and use search/related with item_type: "area".
    • several matches: present the candidates (max 5) and ask once. Never guess. Then read its state (/api/states/<entity_id>) for state, supported_features, source_list, sound_mode_list, group_members, and current media attributes.
  2. Transport / volume / source: check the bit for the exact action first (play 16384, pause 1, stop 4096, next 32, previous 16, seek 2, volume set 4, volume step 1024, mute 8, source 2048, sound mode 65536, turn on 128 / off 256), preview the service + target, then call it: media_player.media_play|media_pause|media_stop|media_next_track|media_previous_track|media_seek (seek_position in seconds; read media_duration/media_position from the state first) |volume_set (volume_level 0.0-1.0) |volume_mute|select_source|select_sound_mode|turn_on|turn_off.
    • Relative volume ("turn it up / down a bit"): that is media_player.volume_up / volume_down, gated by VOLUME_STEP (1024) — a player can support stepping without absolute volume_set (4). Prefer stepping for relative requests, and only fall back to volume_set (read the current volume_level, add a delta) when the player has 4 but not 1024.
  3. Browse: WS {"type":"media_player/browse_media","entity_id":"media_player.<id>"} for the player's own library; drill down by passing media_content_type AND media_content_id together (they are a pair — one without the other is rejected). For Home Assistant's media sources use WS media_source/browse_media, then WS media_source/resolve_media to turn a media-source://... id into a playable URL.
    • Search before drilling when the user names a title, artist, or album: WS {"type":"media_player/search_media","entity_id":"media_player.<id>","search_query":"<text>"} (bit 4194304) searches the player's own library — entity_id is required, and media_content_type/media_content_id scope the search only when supplied together; WS media_source/search_media with search_query (media_filter_classes narrows it) searches Home Assistant's sources. Pass a media_content_id for the source you mean: the media-source ROOT cannot search, so omitting it fails with search_media_failed: Search is not supported for media with content ID even though the payload is valid. Only sources whose can_search is true participate. Both return browse-shaped results you can hand straight to step 4. Neither is universal: when the bit is missing or a source cannot search, say so once and fall back to browsing rather than guessing a content id.
  4. Play a resolved item: media_player.play_media with media_content_id + media_content_type from the browse/resolve step. Never invent a content id.
    • Queue placement uses the same service: enqueue: add | next | play | replace (bit 2097152). "Play this next" is next, "add to the queue" is add; without the bit the player only replaces what is playing — say that instead of silently replacing.
      • add and next have no success signal. They deliberately do not change state or media_title — that is what queueing means — and Home Assistant exposes no standard queue attribute to read back. Step 7's re-read therefore proves nothing here: report the call as accepted, name what would have changed if it had played instead, and do not present unchanged state as a discrepancy.
    • Playback modes are separate services: media_player.shuffle_set (shuffle boolean, bit 32768) and media_player.repeat_set (repeat: off | all | one, bit 262144 — not the grouping bit). "Repeat this song" is one, "repeat the album" is all. Verify these from their OWN attributes — shuffle and repeat in the state — not from the fields step 7 names: a mode change leaves state, volume_level, source and media_title untouched, so the generic re-read would report success without having checked anything.
  5. Grouping (bit 524288 — NOT 262144, which is repeat): media_player.join with group_members: [<entity_ids>] on the master; media_player.unjoin to remove a player. Read group_members back to verify. Speaker-group membership resolves per the shared contract skills/ha-nova/membership-resolution.md — group_members is authoritative, re-read immediately before execution.
  6. TTS announcement: tts.speak with entity_id of a TTS entity, media_player_entity_id of the target, and message. Discover engines with WS {"type":"tts/engine/list"}. For a legacy setup without TTS entities, fall back to the integration's own tts.*_say service and say which one you used.
  7. Verify by re-reading the player's state (state, volume_level, source, media_title) — never report success from the service response alone. Streaming players may show buffering or a briefly unchanged state; wait a few seconds and re-read once before reporting a discrepancy.

Error Handling

Full relay/upstream error taxonomy: skills/ha-nova/relay-api.md -> Error Handling. Media specifics:

  • WS media_player/browse_media fails with a command error when the player does not support browsing — check bit 131072 first.
  • media_content_type and media_content_id must be provided together; a lone id is rejected.
  • A service call can succeed while the player ignores it (offline speaker, wrong source): always verify by re-reading state.
  • media_player/search_media and media_source/search_media exist only on newer Home Assistant versions and only for some integrations or sources — feature-detect (bit 4194304, can_search), and degrade to browsing instead of guessing.

Output Format

Apply skills/ha-nova/output-rules.md to all user-facing output. Write previews, delete confirmations, and results render as the Cards defined there.

Render the Report shape (output-rules.md): what is playing where — player, state, current media, volume, source, and group members when relevant. Browse results render the List Frame (title + type), never the raw tree.

Safety

  • Preview before write: nothing is saved until the user confirms the shown preview.

  • Confirmation binds to the displayed preview and expires on any change to target, payload, endpoint, or scope (context skill → Active Preview Confirmation).

  • Pre-preview phrases ("do it", "go ahead", "implement the plan") authorize drafting and preview only — never the write itself.

  • Delete and destructive operations require the typed confirmation code confirm:<token> verbatim; "yes" or any natural-language reply is invalid.

  • Never guess entity, service, or config IDs — resolve them or ask.

  • Home Assistant is reached exclusively through ha-nova relay.

  • For any HA write this skill does not cover, STOP and invoke ha-nova:fallback first — never probe unfamiliar write endpoints.

  • Drafts follow skills/ha-nova/smallest-solution.md: the complete requested outcome in the simplest safe design, nothing for hypothetical future needs.

  • Volume: preview the target level; treat a large jump (or anything above ~0.8) as disruptive and confirm explicitly — a loud announcement at night is a real-world side effect.

  • Announcements interrupt whatever is playing; say so before sending one.

  • An announcement or TTS to a grouped player plays on every member — name the full member list in the preview, not just the target.

  • Grouping changes affect every listed speaker — preview the full member list.

Guardrails

  • Read supported_features before offering an action; never call a service the player cannot do.
  • One player (or one explicit group) per operation.
  • Never invent media_content_id values — they come from browse/resolve.
  • Browse responses return ALL children of a node — there is no paging parameter. Bound the OUTPUT, not the request: show at most ~20 children per level, say how many more exist, and ask the user to narrow (drill into a folder, name a title) instead of listing on.

Keep looking

Skills are one crate of 627,801. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.