Skip to content

feat(plugins): plugin forms with lists the admin grows, folds, reorders and removes - #2

Open
grtn91 wants to merge 3 commits into
grtn91/request-requester-optionsfrom
grtn91/storage-groups
Open

grtn91 wants to merge 3 commits into
grtn91/request-requester-optionsfrom
grtn91/storage-groups

Conversation

@grtn91

@grtn91 grtn91 commented Nov 11, 2026 •

Copy link
Copy Markdown
Owner

Stacked on the requester options PR.

Problem

Related issue: issue 10 of the Silo-Community Seerr plugin repo (written without a link on purpose, so this fork PR does not show up there)
Validation tasks: none

A plugin's admin form is a fixed set of fields. A plugin whose settings are a list, such as named groups of folders, has to show every entry it could hold as a section of its own. A multi-select also gives no way to see or change the order of its values, which matters when a plugin reads them as a sequence.

Approach

A form section can be one slot of a list (add_group and add_label in the plugin SDK, linked below).

  • The form shows the first slot and every slot up to the last one that holds a value other than its fields' defaults, then a button that reveals the next.
  • A slot folds with a click on its header, using the existing accordion component, also in forms that open their sections.
  • A slot whose first field is a text field shows that value in its header.
  • Arrows in the header trade a slot's values with the slot before or after it, and a delete button removes the slot: the slots after it move up.
  • A multi-select the plugin marks as ordered shows its chosen values as a numbered list that is ordered by dragging, or with the keyboard on the handle. It uses the sortable list hook the other lists use. Options not chosen yet are buttons that add to the end. Other multi-selects keep their toggles.
  • The v2 API passes add_group and add_label through on plugin form sections, and ordered on fields.
  • No new strings: the buttons reuse "Move up", "Move down", "Delete" and "Drag to reorder".

The slots' fields stay ordinary config keys, so stored config and what plugins read do not change. Other collapsible sections keep their Hide/Show button.

Validation

  • Go tests for internal/plugins, internal/requests, internal/api and the plugin, request and contract tests of internal/apiv2: pass.
  • OpenAPI, web types, fixtures, route inventory and local path checks: pass.
  • Web type-check and eslint on the changed files, prettier, and the admin settings and admin component tests: pass.
  • Not run: golangci-lint.

Evidence

Not captured yet. The plugin connection form changes for plugins that declare a list.

Risks

  • go.mod replaces the plugin SDK with a fork commit. It has to become a released SDK version before this can merge.
  • Removing a slot relies on a server's plugin config being saved as a whole, so the emptied slot's keys are dropped.

Checklist

  • I read and can explain the complete diff.
  • This pull request addresses one concern.
  • Each test this pull request adds meets "Decide what to test" in CONTRIBUTING.md.
  • The Evidence section shows every change a user can see, or says there is none.

Related PRs

Based on #1. Same feature in the other repos:

AI Disclosure

  • Harness: Claude Code
  • Tool(s): Claude Code
  • Model(s): claude-fable-5-2
  • Involvement: Fully AI-generated, human verified
  • Adversarial review: n/a

🤖 Generated with Claude Code

…nd reorders

A plugin's admin form is a fixed set of fields. A plugin whose settings
are a list, such as named groups of folders, had to show every entry it
could hold as a section of its own.

A form section can now be one slot of a list (add_group, add_label in
the plugin SDK). The form shows the first slot and every slot up to the
last one that holds a value other than its fields' defaults, then a
button that reveals the next. Each slot folds with a click on its
header, also in forms that open their sections, shows the value of a
leading text field in its header, and has arrows that trade its values
with the slot before or after it.

The v2 API passes add_group and add_label through on plugin form
sections. The slots' fields stay ordinary config keys, so stored config
and what plugins read do not change. Other collapsible sections keep
their Hide/Show button.

The SDK replace in go.mod points at a fork commit until the SDK change
is released.

AI disclosure: written with Claude Fable 5.2 in Claude Code. The plugin
and v2 Go tests, the v2 contract checks and the admin web tests pass
locally. golangci-lint was not run, and the form was not checked in a
browser by the author.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…ove a list slot

Two gaps in plugin forms that hold a list:

A multi-select showed its options as toggles and stored the values in
the order they were clicked, so a plugin that reads them as a sequence
gave the admin no way to see or change that order. A multi-select the
plugin marks as ordered (plugin SDK) now shows its chosen values as a
numbered list. Rows are dragged, or moved with the keyboard on the
handle, with the sortable list hook the other lists use. Options not
chosen yet are buttons that add to the end. Other multi-selects keep
their toggles.

A slot of a list section could be emptied but not removed. Its header
now has a delete button: the slots after it move up and the list shows
one slot fewer.

The v2 API passes the ordered flag through on plugin form fields. No
new strings: the buttons reuse "Delete" and "Drag to reorder".

The SDK replace in go.mod points at a fork commit until the SDK change
is released.

AI disclosure: written with Claude Fable 5.2 in Claude Code. The plugin
and v2 Go tests, the v2 contract checks and the admin web tests pass
locally. golangci-lint was not run, and the form was not checked in a
browser by the author.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
@grtn91 grtn91 changed the title feat(plugins): let a plugin form hold a list the admin grows, folds and reorders feat(plugins): plugin forms with lists the admin grows, folds, reorders and removes Nov 11, 2026
…roups

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
grtn91 pushed a commit that referenced this pull request Nov 15, 2026
…Server#4214)

* fix(audiobooks): match the tagged volume, edition and narrator

Audiobook enrichment accepted a series' first book for a later volume, a
different narrator's edition, box sets as the plain title, and an exact
title alone for items with no author, and it never trusted a tagged ASIN.

Audiobooks now select candidates with stricter rules: a stated volume must
agree, a volume-less series name no longer stands in for volume N, edition
markers must match both ways, an author-less item matches only through an
ID it already holds, and a candidate carrying the tagged ASIN ranks first
unless the titles share no word. Ties resolve the same way whatever order
results arrive in. After a fetch, the record's authors and narrators must
agree with the item (a one-word name only matches itself), and a
narrator-tagged item needs a record that names a narrator unless the record
carries the item's tagged ASIN. A failing candidate is dropped and the next
one fetched, up to three per provider, through the provider gate. A tagged
ASIN is also looked up directly on providers that declare it as a lookup
key. Provider error text is redacted by one helper before it is logged or
stored.

Ebook and manga matching is unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(audiobooks): apply the narrator rule only to the record that establishes a match

A narrator-tagged item needs the record that establishes its match to list
a narrator, unless that record was reached through the item's own tagged
ASIN; the lookup context is now passed in rather than read from the IDs the
provider echoes back. Records merged after the match, including ones
admitted during the merge, are rejected only when they list narrators and
none agree.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(audiobooks): try the next candidate after an empty fetch and reject a stated other volume

- A candidate whose fetch returns no metadata is dropped and the next one
  tried; its IDs and title are kept only once a record passes the checks.
- A fetched record whose title states a different volume than the item's
  title is rejected on every path, the tagged-ASIN lookup included.
- The tagged-ASIN title check also considers the record's aliases.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(audiobooks): treat a number both titles share as part of the title, not a volume

"Slaughterhouse-Five", "Catch-22" and "Fahrenheit 451" were read as volumes
5, 22 and 451, so a record titled "<title>, Book 1" was rejected as another
volume. Audiobook matching now counts an explicit marker ("Book 2", "#2", a
range) as a volume always, and a bare number only when the other title does
not also carry it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(audiobooks): read the trailing number as the volume when a series name carries one

"The 39 Clues 2" and "Area 51 2" were read as volumes 39 and 51; the record
for book 1 repeats that number, so the item ended with no volume and book 1
was accepted. Audiobook matching now reads the last bare number.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(audiobooks): admit a retried tagged-ASIN lookup as a tagged lookup

When a tagged-ASIN lookup fails and the merge retries it with the same
tag-only IDs, the retry now gets the lookup's admission rules, so a record
without narrators is accepted for a narrator-tagged item on either path.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(audiobooks): hold title aliases to the volume rules

An alias that states another volume could carry a candidate's title score:
"Dungeon Crawler Carl 2" accepted "The Gate of the Feral Gods" through its
alias "Dungeon Crawler Carl, Book 4". Aliases now pass the same volume
checks as the primary title before their score counts, and a fetched record
is rejected when any of its titles states a different volume.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(audiobooks): let a record without narrators match, preferring one whose narrators agree

A record that lists no narrators is no longer a contradiction for a
narrator-tagged item. Among one provider's candidates that pass every check,
a record whose narrators agree with the tags is chosen over one that lists
none; the first narrator-less record is used only when no agreeing one is
found. Narrators that disagree still reject a record. The exemptions that
existed only for the strict rule are removed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(audiobooks): reject a record on volume only when none of its titles agrees

A book can sit in two numbered series ("Guards! Guards!" is Discworld 8 and
City Watch 1), so a fetched record is rejected for its volume only when its
titles state another volume and none states the item's.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(audiobooks): accept a narrator-less record as it is; agree on the title that matched

Drop the search for a narrator-agreeing record after a narrator-less one
passed every check: it cost extra fetches against providers that never list
narrators. A record whose narrators disagree is still rejected and the next
candidate tried.

A tagged-ASIN record now sets the cross-provider agreement baseline from the
title or alias that best matches the item, so a translated primary title
does not make a later provider's matching result read as a disagreement.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* refactor(metadata): pick the admitted candidate in one helper

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(audiobooks): try the next candidate after a per-item fetch error; accept another series' numbering

A NotFound or other permanent answer for one candidate's record now moves
on to the next candidate; a paused, throttled or failing provider still
stops for the item, so the gate decides the next call.

Search admission no longer rejects a candidate whose primary title states
another series' number when its original title or an alias states the
item's ("Guards! Guards!: City Watch, Book 1" with the alias
"...Discworld, Book 8"), matching the check after the fetch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(audiobooks): move on only after record-level fetch errors; aliases must name the work

Only a gRPC NotFound or InvalidArgument for a candidate's record moves on
to the next candidate. Provider-wide refusals (Unauthenticated,
PermissionDenied, Unimplemented), untyped errors, throttling and pauses
stop the provider for the item.

An alias may outweigh a primary title in another series' numbering only
when the alias itself passes the title match against the item, not merely
because it carries the item's number.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant