literate-bundles

name

literate-bundles

description

HARD RULE for every OpenTasker JSON bundle (or any JSON) produced for 白い熊 — it must arrive already documented as a "literate programming" artifact: useful per-action labels, a per-task note, and full variable docs at the point of definition. Apply whenever generating/editing an importable bundle (tasks/profiles/variables) or any JSON 白い熊 will read. NOT an opt-in formatting pass — bundles are wrong if shipped uncommented.

Literate bundles — comment everything useful, nothing pointless

白い熊's rule (2026-06-29): anything you produce in JSON must arrive already commented, descriptively and usefully — so a task reads like a literate-programming document. This is not a later polish step; an uncommented bundle is incomplete.

The two anti-goals (hold both at once)

1. Document what isn't self-evident — purpose, units, defaults, ranges/floors, side effects, “needs X permission”, why-this-value. The reader should understand the task from reading it, without asking.
2. Never comment for the sake of it. No labels that just restate the action name/args. If a comment wouldn't tell the reader anything they can't see, omit it — visual flooding is a failure too. Repetitive boilerplate (a run of identical profile.toggle / flow.if / flow.endif) gets one summarizing label on the first, the rest bare.

Where comments live (OpenTasker schema)

• ActionSpec.label (the editor's "Action label", shown first) — the per-action comment. Set it only where it aids legibility (the substantive action, a non-obvious value, a permission requirement).
• itemMeta note (+ noteExpanded: true) — the per-task note. Every task gets one: a one-line literate summary of what the task does / its trigger.
• Project.description — one line on the project's purpose.
• Variables have no description field → document each variable in full on its var.set action label at the point it's defined (usually the …の設定/[01] task): name — meaning, default, range/floor (with the source of any bound, e.g. “floor 40 = ViewConfiguration.getDoubleTapMinTime”), effect. The task's note then points to those action labels as the single source of truth — don't duplicate the prose.

Bilingual (Japanese + English) — required

白い熊 (2026-06-29): every comment is Japanese first, then its English translation on the line(s) below, separated by a blank line. Yes, this doubles the length — that's intentional (白い熊 uses it as a language-learning exercise). Applies to every label, note, and description. Keep the two halves a faithful parallel (same content, not a paraphrase). Format:

<日本語の説明>

<English translation>

(In a JSON string that's "…日本語…\n\nEnglish…".) Skip the English only when the Japanese is itself a bare identifier/value with nothing to translate.

Style

• Comment in Japanese (白い熊's tasks are Japanese), each with the English translation beneath (above rule).
• Keep labels to a tight sentence or two. Put the authoritative detail in one place; reference it elsewhere.
• When folding into an existing task whose style is bare (e.g. 起動完了's unlabeled blocks), stay consistent: update the task note instead of labeling only your inserted lines (which would look odd).
• In the generator, make label/note first-class (e.g. act(type, label=None, **args), im(id, gid, pos, note=None)) so commenting is the default path, not an afterthought.

Ship every item in its group (never orphan it)

白い熊 (2026-07-01): when a bundle carries a task/profile/scene that lives in a group, the bundle MUST carry that group so import files the item into its group — not into "Unfiled". (e.g. the 71/01/37 trio belongs in 起動無効; a bundle shipping task 71 alone dropped it to Unfiled.)

Two parts, both required:

• groups[] — an ItemGroupSpec for each group an included item belongs to (id [bundle-local], tab, projectId, name, position, expanded, parentGroupId). Import MERGES groups by (tab, project, name), so shipping an existing group re-uses it in place (no duplicate, fold state kept).
• itemMeta[] — for each item, its groupId set to that group's bundle-local id (alongside the note). Critical trap: import UPSERTS itemMeta, so it CLOBBERS groupId. Every item already needs an itemMeta row for its note (above) — and an itemMeta emitted without a groupId sets the item's group to null → it jumps to Unfiled on import. So whenever you emit a note, set that item's groupId too.

Source the group + membership from the workspace mirror (_globals/groups.json + _globals/item-meta.json); if the mirror is stale for the item, infer the group by name within its project (match the on-device group). See [[trio-group-convention]].

Reference everything by NAME, never by id (hard rule)

白い熊 (2026-07-04): a bundle references every item by its unique name, never by a numeric id — and the export format is being changed to emit no ids at all. Rules when authoring:

• Never rely on a numeric id matching the device. Ids in a bundle are Room primary keys that leak from storage; on import they're bundle-local and remapped. Setting projectId: 15 and hoping the device's project is #15 silently drops every item to Unfiled (this happened on the 相撲字時計 scenes).
• Always ship projects[] for any project your items belong to, so projectId→project name resolves against the on-device project of that name. The item→project link is name-resolved, not numeric.
• Link by name fields: profile→task via enterTaskName/exitTaskName; scene-element tasks via tapTaskName/longPressTaskName; scene.show/scene.hide by scene name. The numeric twins (enterTaskId, tapTaskId, …) are legacy fallbacks — don't depend on them.
• After the id-free format lands (export writes zero ids, import resolves purely by name), treat any id you emit in a bundle as a bug. See [[reference-by-name-not-id]] and [[bundle-required-fields]] (its "every task needs a unique numeric id" note is superseded once the id-free format ships).

Ship the minimal update set for reimport — not the whole project (hard rule)

白い熊 (2026-07-05): when a change touches only part of a project, ship only the items that changed, never the whole-project bundle by default. Re-importing 154 tasks to fix one scene's xc is waste and noise. But "minimal" means the smallest set that reimports cleanly and dangles nothing — not blindly one item:

• Scene(s) only → ship just the changed scenes. Safe alone: scenes are referenced by NAME (scene.show/scene.hide resolve (project,name)), so overwriting one dangles nothing. (This was the 相撲字時計 time-centre fix — a 6 KB scenes:[semi,unfolded] patch, not the 36 KB project.)
• Task(s) → ship the task plus any profile whose enterTaskName/exitTaskName points at it; re-importing a task alone re-IDs it and dangles the profile link ([[task-overwrite-breaks-profile-link]]).
• Grouped item → still ship its groups[] + the item's itemMeta with groupId (the group rule above), else it jumps to Unfiled.
• Always ship projects[] (name resolution) + the required top-level fields (schemaVersion, appVersion, exportedAtEpochMs) — [[bundle-required-fields]].

Give the generator a flag that emits the minimal slice (e.g. gen_sumo_clock.py --scenes-only) so the narrow patch is reproducible, not hand-carved. The full-project bundle is only the fallback — a first install, or a cross-cutting change that really touches everything.

See [[literate-bundles-rule]], [[minimal-reimport-set]], [[reference-by-name-not-id]], the task-spec skill (action field order incl. Action label), and [[clear-import-instructions]].