P5-search-show: animedex search + animedex show (entity infrastructure + multi-source flagship)

[narugo1992]

Goal

Land Phase 5's flagship entity commands: animedex search <type> <q> and animedex show <type> <prefix:id>. The two commands share one substrate — a prefix-encoded ID parser, a type→backend routing table, a multi-source fan-out helper with per-source fallback, and a common output schema with mandatory source attribution. Both ship in this PR because separating them would duplicate that substrate and force a later reconciliation. After this PR, users have a single command surface that asks "what's this anime called Frieren across every catalogue we cover" and answers it without the user needing to know which backend has what.

Refs #1 §7 (Phase 5 checklist).

Why this slice

search and show are tightly coupled by design:

• They share the type vocabulary (anime / manga / character / person / studio / publisher).
• They share the prefix:id schema — search outputs _source and a backend-native ID; show accepts that backend-native ID with a prefix and routes back to the same backend.
• They share the type → backend routing table — when type is anime, the same set of backends gets queried by search (fan-out) and routed to by show (single-source).
• They share the multi-source fan-out helper — for search, all type-supporting backends are queried in parallel; for show, the prefix selects exactly one backend.

Splitting them into two PRs would mean each PR re-implementing half of this substrate, then merging two divergent implementations. One PR keeps the design coherent.

The aggregate command group is also the first piece of code that introduces the project's flagship multi-source value proposition: animedex search anime "frieren" returning rows from anilist + jikan + kitsu + shikimori, with cross-source merging on rows the matcher confidently identifies as the same anime (shared external IDs first, then a deterministic fuzzy comparison) and source attribution preserved through a sources / records map on every merged row. This commits the project to "wide source coverage with mandatory attribution and intelligent cross-source matching" as the user-facing shape — the merged row is the value, not a step the user has to assemble themselves with jq. Single-source rows that don't match anything on the other backends remain visible with their own attribution.

The merging contract here is the same one P5-calendar (issue #18) establishes for season. Both commands must agree on the merge rule and the merged-row schema; the abstraction proposal step below coordinates that agreement.

Scope (minimum required)

animedex search <type> <q> [--limit N] [--source <backends>]

• <type> ∈ {anime, manga, character, person, studio, publisher}. The positional is required — bare animedex search "frieren" exits with a clear "missing TYPE" Click error, not a guess.
• <q> is the search string, passed verbatim to each upstream's search endpoint.
• --limit N is per source (default 10). The merge step runs after the per-source limit, so a --limit 5 against 4 sources can return between 5 and 20 merged rows depending on overlap. To globally cap, use --jq 'limit(N; .)' downstream.
• --source <comma-separated> collapses the fan-out to a subset of backends. E.g. --source anilist,jikan skips kitsu / shikimori.

The type → backend mapping (minimum required surface, derived from the high-level helpers already shipped on main):

Type	Backends fanned out by default
anime	anilist, jikan, kitsu, shikimori
manga	anilist, jikan, kitsu, mangadex, shikimori
character	anilist, jikan, kitsu, shikimori
person	anilist, jikan, kitsu, shikimori
studio	anilist, jikan, kitsu, shikimori
publisher	shikimori (only backend with a publisher catalogue today)

If a backend in the fan-out fails (rate-limit, 5xx, network), the fallback policy (same as P5-calendar) applies: per-source try/except, the successful sources' rows still come back, the failed sources surface in an envelope sources: {...} map with {status, reason, http_status, message, duration_ms}. Stderr emits one inform line per failed source. Exit code 0 if at least one source succeeded; non-zero only when every source failed.

animedex show <type> <prefix:id>

• <type> is required, same vocabulary as search.
• <prefix:id> is the project's canonical entity reference. Examples:
  • anilist:154587
  • mal:52991 (alias for jikan, because Jikan wraps MyAnimeList)
  • kitsu:43534
  • shikimori:52991
  • mangadex:dc8bbc4c-eb7a-4d27-b96a-9aa8c8db4adb
• Single-source dispatch: the prefix picks exactly one backend. The command routes to that backend's show() / manga_show() / character() / etc. helper per the type.
• Bad combinations (e.g. show publisher anilist:1 — anilist has no publisher catalogue) exit with a clear "type 'publisher' is not supported by backend 'anilist'" error and a one-line list of supported backends for that type.

Prefix → backend mapping (minimum required)

Prefix	Routes to backend module	Notes
anilist	animedex.backends.anilist	numeric ID
mal	animedex.backends.jikan	numeric ID; alias for jikan
jikan	animedex.backends.jikan	numeric ID; same as mal
kitsu	animedex.backends.kitsu	numeric ID
shikimori	animedex.backends.shikimori	numeric ID
mangadex	animedex.backends.mangadex	UUID string
ann	animedex.backends.ann	numeric ID

Treat anidb: as a documented-but-deferred prefix that raises a typed informative error pointing at Phase 7. Do not silently swallow anidb: references.

Encouraged exploration

The minimum surface above is the floor. AGENTS §15.2 ("read-only API surface coverage") asks for everything the upstreams support that fits the pattern, not the cherry-picked subset. Encouraged additions:

• More type entries: topic (shikimori), club (shikimori), report (ann) — only if the high-level helper exists and the upstream's search returns a clean list shape. Topics / clubs are community resources rather than canonical entities; if they look out of place under search anime, that's the right reason to keep them as future work.
• Reverse-prefix aliases: myanimelist: as a long-form alias for mal:, etc. Cheap and friendly.
• A --first flag that returns only the first result row per source (useful for piping into show).
• Concurrent fan-out (concurrent.futures.ThreadPoolExecutor(max_workers=len(sources))) so a 4-source search anime completes in max-of-4 time rather than sum-of-4 time. Default to concurrent if the abstraction proposal sign-off accepts it.

API documentation entry points

The high-level helpers this PR composes already exist on main. The relevant module heads:

• AniList: animedex/backends/anilist/__init__.py — search() / show() / character() / staff() / studio().
• Jikan: animedex/backends/jikan/__init__.py — same shapes. Note Jikan's staff is per-anime; the top-level person endpoint is person().
• Kitsu: animedex/backends/kitsu/__init__.py.
• Shikimori: animedex/backends/shikimori/__init__.py — already exposes search, show, manga_search, manga_show, ranobe_*, club_*, publishers, people_search, person, characters, staff per PR dev(narugo1992): add shikimori and ann high-level backends #15.
• MangaDex: animedex/backends/mangadex/__init__.py — for type=manga fan-out, this is the manga-native catalogue (uses UUID IDs, not numeric).
• ANN: animedex/backends/ann/__init__.py — search() returns multiple anime; show() accepts numeric ID.

If a helper is missing for a particular type/backend combination listed in the type table above, that's a real gap and ought to be filled in this PR (per §15.2). Most of the gaps were closed in PR #15's shikimori expansion; spot-check before assuming the helper exists.

Substrate touch points (read carefully)

This PR introduces the project's first aggregate-shape layer. Before writing implementation code, post an abstraction proposal as a comment on this issue and wait for an explicit external sign-off — sign-off is a recorded maintainer reply, not a self-reply (AGENTS §15.5; see the merged PR #14 issue-#11 comment and the merged PR #15 issue-#12 comment for the well-formed proposal shapes; both received an explicit maintainer "+1" before implementation began).

The proposal should answer at least:

1. Where does aggregate code live? Recommended: animedex/agg/ as a new sibling to animedex/backends/. The module split inside:

   • animedex/agg/_prefix_id.py — parse(prefix_id: str) -> (backend: str, id: str) and reverse; also the prefix → backend module name table.
   • animedex/agg/_type_routes.py — backends_for_type(type: str) -> List[str] and the type vocabulary.
   • animedex/agg/_fanout.py — the multi-source fan-out helper. (If P5-calendar lands first and puts this helper in the same place, this PR reuses it; merge the proposal accordingly.)
   • animedex/agg/search.py, animedex/agg/show.py — the high-level command implementations.
   • animedex/entry/search.py, animedex/entry/show.py — the Click entry modules.

2. The output schema for search. Each row carries the rich-model dump from its originating backend (e.g. a row from anilist is the full AnilistAnime.model_dump(by_alias=True), not a lowest-common-denominator projection). The row also carries a top-level _source (matching the existing source_tag convention) and a _prefix_id (e.g. anilist:154587) so the user can feed search output directly into show. Confirm the rich-model-preserved approach in the proposal.

3. Common-model projection for TTY rendering. JSON output keeps the full rich row. TTY output renders a compact table (title / score / status / [src:]) using to_common() projections for display. Confirm the renderer path; this is the only place lossy projection is acceptable.

4. Fan-out strategy: concurrent vs sequential. Up to 5 backends per search call (manga case: 5 sources). Concurrent (ThreadPoolExecutor) is the natural choice; rate-limit buckets are per-backend so there is no contention; the existing transport stack is thread-safe via requests.Session per call. Propose the default with rationale.

5. show with unsupported (type, backend) combinations. E.g. show publisher anilist:1. Two options: (a) exit with a Click error before any network call; (b) attempt the upstream and return whatever 404 it gives. (a) is more informative — propose it. The error message should name the type, the offending backend, and the list of backends that do support the type.

6. Conflict with P5-calendar's _fanout.py. If P5-calendar merges first, this PR consumes its animedex/agg/_fanout.py verbatim. If this PR merges first, P5-calendar consumes ours. The proposal step coordinates which PR's fan-out helper definition is canonical; the helper itself should be generic (no search-specific assumptions baked in).

7. anidb: prefix handling. The advisory classifier should recognise the anidb: prefix and raise a ApiError(reason="auth-required") or a new reason like "deferred-feature" with a message such as "ANN ANIDB / AniDB high-level helpers are not yet shipped; use animedex api anidb /... to call the raw passthrough once it lands". Settle the reason string in the proposal.

If the proposal you write diverges materially from these defaults, that is fine; the proposal exists to surface the trade-off in writing before code lands.

Fixture capture

Fixture corpora for each fan-out source already exist on main (see test/fixtures/<backend>/). For search and show:

• Happy path — reuse existing per-backend search/show fixtures. For search anime "frieren", expect to compose fixtures from test/fixtures/anilist/anime_search/01-frieren.yaml, test/fixtures/jikan/anime_search/01-frieren-52991.yaml (or equivalent), test/fixtures/kitsu/anime_search/01-frieren.yaml, test/fixtures/shikimori/animes_search/01-frieren.yaml. Spot-check each exists; if a probe is missing for a type+backend combo, capture one as part of this PR using tools/fixtures/run_<backend>.py.
• Partial failure — synthetic 429 / 5xx fixtures using the same hand-edit technique as P5-calendar (edit a captured fixture's response.status and add response.captured_from: synthetic-429 metadata).
• Total failure — synthetic 5xx across every fan-out source for one type.

show reuses single-source per-backend show fixtures; no fan-out fixture needed except for the prefix-parser tests.

Document capture date and synthetic-fixture notes in the PR body (AGENTS §15.5).

Verification checklist (self-check before requesting review)

Search

• animedex search anime "frieren" fans out to all 4 anime-supporting backends; envelope contains rows from each, every row has _source and _prefix_id.
• animedex search manga "berserk" fans out to all 5 manga-supporting backends including mangadex.
• animedex search character "frieren" fans out to all 4 character-supporting backends.
• animedex search person "miyazaki" fans out to all 4 person-supporting backends.
• animedex search studio "ghibli" fans out to all 4 studio-supporting backends.
• animedex search publisher "kodansha" is single-source on shikimori; envelope's sources map contains only shikimori.
• animedex search (no type) exits with a Click missing-argument error mentioning the supported types.
• animedex search badtype "x" exits with a clear "unknown type" error listing valid types.
• --source anilist,jikan collapses the fan-out to the named subset; other backends do not appear in the sources map.
• --limit 5 returns at most 5 rows per source.
• Partial failure: synthetic-429 fixture for anilist + healthy fixtures for jikan/kitsu/shikimori; the command exits 0, returns rows from the 3 healthy sources, stderr names anilist's failure.
• Total failure: synthetic 5xx for every fan-out source; command exits non-zero, stdout is an empty-items envelope with sources showing every status as failed.

Show

• animedex show anime anilist:154587 routes to anilist.show and returns the rich anilist record with source attribution.
• animedex show anime mal:52991 routes to jikan.show(52991).
• animedex show anime jikan:52991 also routes to jikan (mal/jikan alias).
• animedex show manga mangadex:<uuid> accepts the UUID format.
• animedex show character anilist:206439 routes to anilist.character.
• animedex show person shikimori:1870 routes to shikimori.person.
• animedex show studio anilist:21 routes to anilist.studio.
• animedex show publisher anilist:1 exits with a clear "type 'publisher' is not supported by backend 'anilist'; supported backends: shikimori".
• animedex show anime badprefix:1 exits with a clear "unknown prefix" error.
• animedex show anime anidb:42 exits with the Phase-7-deferred informative error message.
• animedex show anime anilist:abc exits with a clear "ID is not numeric for backend 'anilist'" error before any HTTP call.
• animedex show anime anilist:9999999999 (valid format, 404 upstream) propagates the upstream 404 cleanly.

Cross-cutting

• CLI tested in both --json and the default TTY path with isatty()=True forced (AGENTS §9bis.6).
• HTTP is the only mock seam — responses.RequestsMock against captured/synthetic fixtures; no monkeypatch.setattr(animedex.backends.<x>, ...) above-the-wire shortcuts (§9bis.1).
• No new entry in animedex/transport/read_only.py — these commands compose existing backends, not new ones.
• _BACKEND_POLICY in animedex/entry/_cli_factory.py gains entries for the two aggregate commands if registered as top-level Click groups; \f cutoff convention from §10 step 6 honoured.
• make rst_auto regenerates docs/source/api_doc/agg/ and the diff is committed.
• Tutorial entries at docs/source/tutorials/aggregate.rst (or split per command) with at least one runnable search example, one show example, and one example showing the partial-failure shape.
• _SELFTEST_TARGETS registers every new animedex.agg.* module.
• Pre-flight (AGENTS §15.4): full sequence green; grep -rE 'Phase [0-9]|AGENTS[. ]§|Reviewer review' animedex/ tools/ returns zero.

Load-bearing reminders

• Lossless rich model under merging — when search groups multiple backends' rows into one merged entry, each upstream's full rich-model dump is preserved under the merged row's sources / records map. §13 still applies: no upstream-visible field disappears. The merged row's compact TTY rendering may project a common shape for the eye, but the JSON path keeps each upstream's full rich record. Both score values stay visible under the merged row (Phase 5 contract: do not average; surface both).
• Cross-source merging is the user-visible default, not an opt-in. Matching is conservative: external IDs first, then a deterministic fuzzy comparison; a single threshold cutoff determines when two upstreams' rows are merged into one entry. Rows that do not meet the threshold remain as their own single-source entries — they are not silently dropped. The merge is what the multi-source search value exists to produce.
• §0 inform-not-gate on degraded sources — failed sources never crash the command. The user always gets the data the healthy sources returned; the failed sources surface as stderr inform lines and a structured sources envelope entry.
• §15.2 read-only surface coverage — type → backend mapping covers every backend that genuinely has the entity. Spot-check before deferring a backend; "I didn't have time" is not a valid reason to drop a backend from the table.
• HTTP-only mock seam — tests load real captured fixtures (and the documented synthetic 429/5xx fixtures) through responses.RequestsMock. No project-internal monkeypatch.setattr.

Parallelism

This issue can be picked up by codex while the P5-calendar issue runs in parallel. They share zero source files; the only conflict point is the registration line in animedex/entry/__init__.py (or wherever top-level Click commands are wired) where each PR adds its own subcommand — a trivial textual merge.

The fan-out helper at animedex/agg/_fanout.py is contributed by whichever PR merges first. The abstraction proposal step on both issues should reach agreement on the helper's signature before either PR writes code; if there is uncertainty, the first-to-merge PR's shape is canonical and the second PR refactors to consume it.

The P5-crossref issue depends on this PR's _prefix_id.py parser and waits for this PR to merge.

PR body template (what to include)

When you open the PR, please include:

• Summary — one-paragraph overview of the new commands, the substrate they introduced (prefix:id parser, type routes, fan-out helper), the fallback contract.
• Demos — TTY GIFs for both search and show. For search, ideally a captured run that shows multi-source attribution (rows from 4 sources with [src:] tags). For show, demonstrate one anime + one character + one studio command in the same recording.
• Examples and expected output — copy-pastable command lines + jq-extracted expected output. Include one partial-failure example.
• Fixture notes — capture date in UTC; explicit synthetic-fixture disclosure; proxy notes if any.
• Verification — tick the checklist; for each item write done / partial / deferred with a one-line reason.
• Abstraction proposal link — link to the comment on this issue where you settled the substrate design, plus the reviewer comment that signed off (mandatory per §15.5).

Out of scope

• --no-merge / inspect-raw-fan-out toggle (not required; merged rows keep per-upstream rich records under the sources / records map, so a caller who needs the raw upstream rows reads them from there).
• Bare animedex search "frieren" without explicit type — by design.
• animedex show <prefix:id> without type — by design (same ID space can encode anime or manga on most backends).
• animedex crossref — that's P5-crossref.
• AniDB integration for anidb: prefix — Phase 7.
• Calendar commands (season, schedule) — P5-calendar.
• Filters like --genre, --year, --studio on search — future extension.

[narugo1992]

Proposed aggregate-layer shape for #19:

1. Put the new code under animedex/agg/ as a sibling to animedex/backends/, with _prefix_id.py, _type_routes.py, _fanout.py, search.py, and show.py.
2. Keep search rows lossless: each returned row stays the originating backend's rich-model dump, with _source and _prefix_id attached. No cross-source dedupe, no score averaging.
3. Keep TTY rendering compact and lossy only there: TTY should display a common-model projection for readability, while JSON preserves the full rich row.
4. Use concurrent fan-out by default with ThreadPoolExecutor, since the backends are independent and rate limits are per backend.
5. Make show validate the (type, backend) pair before any network call. Unsupported combinations should fail early with a clear Click error that lists the supported backends for that type.
6. Treat anidb: as a deferred prefix that raises a typed informative error instead of silently falling through.
7. Reuse the same generic _fanout.py shape for any future calendar-like aggregate command so the helper stays shared rather than search-specific.

If this matches the issue intent, I will implement it that way and keep the PR body explicit about the fallback envelope for partial source failures.

[narugo1992]

Implementation PR opened: #22

Current status:

• Added animedex search <type> <q> and animedex show <type> <prefix:id>.
• Added the shared aggregate substrate under animedex/agg/ plus animedex.models.aggregate.
• Kept rows lossless in JSON with _source and _prefix_id; TTY uses compact source-attributed rows.
• Implemented partial-failure semantics: valid stdout envelope, stderr inform lines per failed source, exit 0 when at least one source succeeds, non-zero when every selected source fails.
• Covered search/show with HTTP fixture replay only; no backend/helper monkeypatching above the HTTP seam.
• Added aggregate tutorial, API docs, fixture prewarm helper, and aggregate.gif demo.

Verification already completed locally:

• make format
• make format-check
• python -m animedex.policy.lint animedex/
• python -m animedex --help
• python -m animedex search --help
• python -m animedex show --help
• python -m animedex selftest (111 passed, 0 failed)
• targeted aggregate/entry/render tests (70 passed)
• make test (2810 passed, 84 skipped, total coverage 99%)
• make rst_auto
• make docs (warning-free)
• make build && make test_cli (4 passed, 0 failed)
• git diff --check

Fixture/source notes are documented in the PR body, including the synthetic ANN 503 fixture and the observed temporary/shape limitations from AniList typed search and Kitsu people search.

[narugo1992]

Coordination note on the repository English-only rule after the working-thread clarification.

The rule is about contributor-authored repository prose and interface text: source comments, docstrings, generated docs, examples, PR/issue bodies, command help, commit messages, and other human-authored explanatory text must remain English.

It does not require factual data to be translated or replaced. Fixture payloads, captured upstream response values, entity titles, names, slugs, user queries, JSON literals, and test strings that represent real upstream or user data may keep their original language and non-ASCII characters when that is the correct data. In particular, #19 search/show tests and fixtures should preserve real titles/names from upstream data rather than romanizing or fabricating English-only substitutes just to satisfy the prose rule.

I updated AGENTS.md on the PR branch to make that distinction explicit.

[narugo1992]

Coordination update after #18 / PR #21.

The calendar aggregate slice has been squash-merged into main as 6a15de2. The shared pieces that #19 should now consume or stay compatible with are:

• animedex/agg/_fanout.py for multi-source fan-out and partial-failure capture.
• animedex/models/aggregate.py for the aggregate envelope/status models.
• The partial-failure contract already on main: stdout remains structurally valid, healthy sources still return rows, stderr reports each failed source, exit code is 0 when at least one selected source succeeds, and non-zero only when every selected source fails.
• The season merge implementation also established the conservative merged-row provenance pattern: compact core fields for display, with per-source records/details/payloads preserved for JSON consumers.

This does not close #19. search, show, and crossref remain open in #1; #19 should continue against the merged aggregate substrate rather than carrying a divergent helper/model shape.

[narugo1992]

Scope clarification after PR #21 (P5-calendar) merge

PR #21 has merged the cross-source merge substrate (MergedAnime / _merge_season_items / _anime_match_score / id_conflicts / merge_diagnostics / etc.). PR #22 was opened before PR #21 merged, so its current head implements search as annotate-only (_source + _prefix_id per row, no merging) on its own version of _fanout.py / AggregateResult. The shape this issue's spec now describes — "cross-source merging is the user-visible default" — is the eventual target, not what PR #22 should ship in its current round.

The decision (maintainer-confirmed): this round ships search annotate-only; cross-source merging on search lands in a follow-up slice that reuses PR #21's substrate directly. That keeps PR #22 from doubling in scope while it is already mid-stream and lets the merge work land against the now-stable _merge_season_items shape rather than against a parallel implementation.

What PR #22 should do this round

1. Merge main into the branch first. PR P5-search-show: add aggregate search and show commands #22's baseline is from before PR Add calendar aggregate commands #21, so main now carries substrate this PR overlaps with. See PR P5-search-show: add aggregate search and show commands #22 for the concrete file-by-file merge guidance and the post-merge verification checklist.
2. Adopt PR Add calendar aggregate commands #21's substrate verbatim — animedex/agg/_fanout.py, animedex/models/aggregate.py, animedex/entry/aggregate.py, animedex/utils/timezone.py, the requirements.txt additions (anyascii, jaconv, unidecode, tzdata, python-dateutil), and the tools/generate_spec.py hidden-imports. PR P5-search-show: add aggregate search and show commands #22's parallel implementations of these are pre-PR-Add calendar aggregate commands #21 work and should be discarded in favour of main's versions.
3. Land search and show against that substrate, with search returning a flat list of rows annotated with _source and _prefix_id. The same anime appearing in multiple backends comes back as multiple rows — that is acceptable for this slice. Source attribution is mandatory on every row.
4. Use the shared CLI helpers in animedex/entry/aggregate.py (_report_failures / _report_merge_diagnostics / _finish / _emit) rather than duplicating them in entry/search.py / entry/show.py. Register both commands as direct subcommands of animedex_cli, not as pseudo-backends under _BACKEND_POLICY (per the spec above).
5. Defer cross-source merging on search to a follow-up issue/PR. The merge logic from PR Add calendar aggregate commands #21 (_anime_match_score, _external_id_conflicts, _choose_merged_title, the calibrated _MERGE_THRESHOLD and adjudication corpus) is anime-specific by design; extending it to manga / character / person / studio / publisher will need either a per-entity match-score function or a generic interface, plus its own adjudication corpus for the entity types beyond anime. That work is its own PR with its own abstraction proposal.

Follow-up slice (separate issue, to file after PR #22 merges)

When this round merges, a new follow-up issue will land that:

• Refactors _merge_season_items and _anime_match_score into a generic _merge_aggregate_items(items, *, match_score=...) helper plus per-entity scorers (_anime_match_score, _character_match_score, _studio_match_score, etc.) where each entity type ships its own threshold and adjudication corpus.
• Wires search to call that merge helper by default, mirroring season's shape.
• Updates this issue's spec from "annotate-only" to "intelligent cross-source matching" as the closed user-facing contract.

Spec wording for this round

Bullets in the "Output schema" / "No cross-source dedupe" sections of this issue describe the eventual state (post-follow-up). For this round's PR review, treat the contract as:

• search rows are annotated with _source (matching the existing source-attribution convention) and _prefix_id (a stable <backend>:<id> string the show command can consume directly).
• Same anime in multiple backends returns as multiple rows. No merge in this slice.
• --limit N is per source; --source collapses fan-out.
• Partial-failure semantics from PR Add calendar aggregate commands #21 apply: stderr inform line per failed source, exit 0 if at least one source succeeded, non-zero only when every selected source failed.
• _BACKEND_POLICY is not modified; search/show are top-level commands.

After PR #22 merges, this issue stays open with its scope re-pointed at the follow-up merge slice, and [search annotate-only] / [show] get ticked on issue #1 §7.

[narugo1992]

This issue is closed automatically by the merge of PR #22 (commit 138281ef). Recording what landed and what was scoped out for the record:

Landed in PR #22

• animedex search <type> <q> — multi-source fan-out across every type-supporting backend, annotate-only (one row per backend, with _source and _prefix_id).
• animedex show <type> <prefix:id> — prefix-encoded ID routing back to the owning backend.
• Aggregate substrate carry-overs from PR Add calendar aggregate commands #21 (animedex/agg/_fanout.py, animedex/models/aggregate.py) consumed verbatim.
• Type → backend routing for anime / manga / character / person / studio / publisher.
• Partial-failure semantics: per-source try/except, stderr inform lines per failed source, exit 0 if at least one source succeeded.

Deferred to follow-up #23

• Cross-source merging on search — agreed during the PR scope clarification to ship as a separate slice. PR P5-search-show: add aggregate search and show commands #22 stayed annotate-only.
• A new animedex enhance <query> stateless suggestion command for agent workflows.

Deferred to a separate small bug-fix PR (not bundled in #23)

• JikanPerson.alternate_names rejects null (real upstream null in ~4% of person rows).
• Kitsu _fetch misreports a 400 ("Filter not allowed") as upstream-shape: missing 'data' key.

Both were surfaced by the empirical eval posted on PR #22 (issuecomment-4439876211) and should land independently of the merge work.

#1 status block updated in a separate comment there. #20 (P5-crossref) is now unblocked — the _prefix_id.py parser it depends on is on main.

[narugo1992]

Follow-up issue for the deferred search-merge work and the new animedex enhance agent-composable layer is #23.