{"entries":[{"lastCommit":{"author":{"email":"brendan.schlagel@gmail.com","name":"Brendan Schlagel","when":"2026-07-10T17:23:52.000000-04:00"},"committer":{"email":"noreply@github.com","name":"GitHub","when":"2026-07-10T17:23:52.000000-04:00"},"hash":"77d5fdf7b15da57e7b61fd40d272837b997b7537","message":"Merge pull request #349 from hyperlink-academy/feature/pubBlock\n\nFeature/pub block","tree":"e24f816201631bb6d744647573edbd344bf055ed"},"mode":"0100644","name":"attributes.ts","size":11721},{"lastCommit":{"author":{"email":"jared@awarm.space","name":"Jared Pereira","when":"2026-07-20T15:27:19.000000-04:00"},"committer":{"email":"jared@awarm.space","name":"Jared Pereira","when":"2026-07-20T15:30:49.000000-04:00"},"hash":"3c851dd284723f6e05f12fbd3ba26df27e2d4c90","message":"cached mutation context handle updates as well as del/write\n","tree":"e2ae6db8d6f0fef6a99e78f5d7f817ffc5ec75ab"},"mode":"0100644","name":"cachedServerMutationContext.ts","size":11524},{"lastCommit":{"author":{"email":"jared@awarm.space","name":"Jared Pereira","when":"2026-06-16T09:53:30.000000-07:00"},"committer":{"email":"noreply@github.com","name":"GitHub","when":"2026-06-16T12:53:30.000000-04:00"},"hash":"3b3602f391d64ab759ac663a4a3459056dc84ac8","message":"Refactor undo/redo to support coalescing text edits (#321)\n\n* docs: add text editor behaviors reference\n\nStart a living doc of expected text editor behaviors. First entry: Enter\non a blank list item outdents one level, and at the top level converts the\nitem to a normal non-list block while preserving its text type (heading,\nblockquote, or plain text).\n\nhttps://claude.ai/code/session_01V4wW9QQpVzduvU5K3BzNGq\n\n* docs: drop implementation pointers from editor behaviors\n\nKeep the doc a purely human-readable specification of behavior.\n\nhttps://claude.ai/code/session_01V4wW9QQpVzduvU5K3BzNGq\n\n* Fix list Enter races in creation and outdent\n\nBoth the create and outdent paths drove structural mutations off React-prop\nsnapshots (which lag the Replicache store) and a fixed focus timer, so fast\n\"type, Enter, type, Enter\" or repeated Enter-on-blank produced lost focus,\ncolliding positions, transiently non-list items, and dropped blocks.\n\nCreate path (keymap `enter`):\n- Position the new sibling from the freshly-read sibling rather than the\n stale `propsRef.current.position`, so it can't collide with or sort before\n the item just created.\n- Create the list item atomically: read inherited list-style/check-list up\n front and write type + is-list + style + checklist in a single `addBlock`\n mutation, closing the window where the new block wasn't yet a list item and\n a follow-up Enter inserted a plain paragraph.\n- Focus when the new block's editor actually registers (retry) instead of a\n fixed 10ms timeout that silently dropped focus, sending keystrokes to the\n previous block.\n\nOutdent path (`outdentBlock`):\n- Validate the destination anchor before any retraction. Previously a\n stale/missing `after` returned after the block was already retracted and its\n following siblings re-parented under it, orphaning the block and everything\n below it.\n\nhttps://claude.ai/code/session_01V4wW9QQpVzduvU5K3BzNGq\n\n* Trim list-Enter fix comments to rationale only\n\nDrop comments that just narrate what the code shows; keep the non-obvious\nwhy (atomic creation avoids a non-list window, validate-before-retract avoids\norphaning, focusBlock no-ops until mount).\n\nhttps://claude.ai/code/session_01V4wW9QQpVzduvU5K3BzNGq\n\n* Group multi-mutation actions into single undo steps\n\nUndo entries are added asynchronously inside Replicache mutator bodies, so a\nuser action collapses to one undo step only when its handler opens a group,\nawaits every mutation, then closes it. Many handlers fired multiple mutations\nungrouped (or fire-and-forget), fragmenting one action into several Cmd-Z steps.\n\n- Self-group shared helpers via an optional undoManager param, awaiting their\n internal mutations: list-operations indent/outdent/outdentFull/\n multiSelectOutdent (indent previously fire-and-forgot, breaking grouping even\n on the keyboard path) and deleteBlock (which called a bare endGroup with no\n matching start). Ungrouped callers thread undoManager; already-grouped callers\n rely on the global isGrouping flag.\n- BlockCommands await createBlockWithType + facts; BlockCommandBar groups both\n the keyboard and click onSelect paths via throw-safe withUndoGroup.\n- TooltipButton uses withUndoGroup; ToolbarButton awaits its onClick so the\n group spans the async work (onClick typed => unknown to keep void leniency).\n- addLinkBlock gains ignoreUndo (mirroring addImage) so async link enrichment\n after a paste doesn't add stray undo steps.\n- Wrap remaining leaf handlers: block submits, text-type toolbar, mention\n embed, image upload, canvas/doc drops, theme/image pickers, list toolbar,\n block swipe, poll save, posts-list tag filters.\n\ncreateBlockFromHTMLLegacy (legacy/canvas paste, marked for deletion) is left\nungrouped pending a full async rewrite.\n\nCo-Authored-By: Claude Opus 4.8 (1M context) \n\n* Fix undo losing the block after indent\n\nindent fired retractFact + addLastBlock without awaiting them. main's undo\nrefactor closes the undo group when the command promise settles (replacing the\nold 100ms endGroup timeout), so the un-awaited mutators registered their undo\nops after the group closed — as two separate entries. One undo then reversed\nonly addLastBlock (retracting the new card/block fact) and orphaned the block;\na second undo restored it.\n\nAwait both mutations (as outdent already does) so their undo ops land inside\nthe group and a single undo restores the block to its prior position.\n\nhttps://claude.ai/code/session_01V4wW9QQpVzduvU5K3BzNGq\n\n* Fix undo/redo data loss and position swaps in list ops\n\nThree independent defects surfaced by the subagent investigation:\n\n- retractFact's redo closure omitted ignoreUndo, so redoing any indent/outdent\n re-entered the undo manager — truncating the redo stack and pushing a dangling\n entry that double-referenced then lost the block. Add ignoreUndo (every other\n closure already passes it).\n\n- indent did retract-old-fact + add-new-factID. A mis-grouped undo could apply\n only the add and orphan the block. Reparent the existing card/block fact in a\n single mutation (reuse its factID) so it's one symmetric undo entry that can't\n be half-applied.\n\n- the nested sibling sort in getBlocks had no id tiebreak (the top-level sorts\n do), so equal/colliding positions rendered in unstable order. Add the tiebreak.\n\nhttps://claude.ai/code/session_01V4wW9QQpVzduvU5K3BzNGq\n\n* Rework undo grouping and keep focus across reparenting\n\nGrouping: a single shared isGrouping flag was driven by three producers — Tab's\nwithUndoGroup, the keymap Enter/Backspace groups, and a 200ms text-edit timeout\ngroup in trackUndoRedo — that closed each other's groups and split a command's\nmutations across boundaries (orphaning blocks on undo) or merged a Tab into\nadjacent typing. Centralize all grouping in the undo manager:\n- command groups are depth-counted (nesting-safe) and flush any open text-edit\n group first, so a command never merges with typing;\n- text edits coalesce through addGrouped/flushGroup (the manager owns the 200ms\n window), so trackUndoRedo no longer races a per-component timer;\n- undo/redo are serialized so rapid keypresses don't re-enter @rocicorp/undo's\n recursive group walk and corrupt the stack.\n\nFocus: indent/outdent change a block's parent but keep its entityID. parent was\na dependency of the ProseMirror mount effect, so every structural edit (and its\nundo/redo) destroyed and recreated the EditorView, dropping the caret to .\nDrop parent from the dependency array (it's read via propsRef) so the view\nsurvives reparenting and the caret follows the block.\n\nhttps://claude.ai/code/session_01V4wW9QQpVzduvU5K3BzNGq\n\n* Coalesce text-edit undo by span key, not just time\n\naddGrouped coalesced every text edit within the 200ms window regardless of\nsource, so edits to two different blocks within the window merged into one undo\nstep (only the blur flush saved it). Take a span key (the block/footnote\nentityID): consecutive same-key edits coalesce within the window, and a\ndifferent key finalizes the open group immediately so distinct spans never\nmerge — independent of timing or blur. trackUndoRedo threads the editor's\nentityID through as the key.\n\nhttps://claude.ai/code/session_01V4wW9QQpVzduvU5K3BzNGq\n\n* Stop splitting typing runs in undo coalescing\n\nText-edit coalescing was correct, but the group got finalized between\nkeystrokes, so each character became its own undo step:\n\n- The

 onBlur handler called undoManager.flushGroup() on every blur, and\n  focus churn during the deferred-updateState re-render fired it mid-run.\n  Remove the blur flush (and the now-unused flushGroup); the idle timer, command\n  groups, span-key changes, and undo/redo already provide every real boundary.\n- COALESCE_MS was 200ms — shorter than a normal inter-keystroke gap (typing\n  below ~5 cps), so the timer flushed between keystrokes. Raise to 500ms; the\n  timer resets each keystroke, so a run now coalesces until an actual pause.\n\nhttps://claude.ai/code/session_01V4wW9QQpVzduvU5K3BzNGq\n\n* Fix typing-run undo coalescing broken by per-keystroke command groups\n\nTyping a continuous run was undoing/redoing one character at a time instead\nof as a single step.\n\nRoot cause: SelectionManager registers a global window keydown listener that\nwraps its whole body in undoManager.withUndoGroup(...) and only checks which\nkey it handles inside the callback. So every keystroke — including plain\ncharacters it ignores — opened a command group. On this branch startGroup()\neagerly called flushCoalesce() and opened a raw @rocicorp/undo group, so each\nkeystroke finalized the in-progress text-edit coalescing group before the\ncharacter's own edit was recorded. Every character therefore landed in its own\nundo group. (On main, withUndoGroup was a no-op when a group was already open,\nso it never disturbed the coalescing run.)\n\nFix: make command groups lazy. startGroup() now only records intent\n(groupDepth++); the raw group is opened — and adjacent typing flushed — lazily\nin openCommandGroup() on the command's first recorded entry (via add()). A\nno-op keystroke wrapper records nothing, so it leaves the typing run intact; a\nreal command still flushes the run and becomes its own undo step. addGrouped()\nkeys off whether a command group is actually materialized rather than merely\npending, so an externalUndoGroup edit (e.g. the Enter split) still joins its\ncommand while plain typing keeps coalescing.\n\nVerified against @rocicorp/undo@0.2.1: a continuous run shares one groupID and\none undo reverts it; type/command/type yields three distinct steps.\n\nhttps://claude.ai/code/session_017QZrKm9paHWqRQ1yMsSty2\n\n* Fix block disappearing when undoing/redoing a move\n\nmoveBlock, moveBlockUp, and moveBlockDown each moved a block by retracting its\ncard/block fact and then asserting a new one with the same id. That produced two\nundo entries with an intermediate state in which the fact does not exist, so\nundoing (or redoing) a move blanked the block: one Cmd-Z removed the block, and\nonly a second restored it at the old position.\n\nDrop the redundant retract and let the single assertFact overwrite the existing\nfact in place (reusing its id) — the same single-mutation pattern addLastBlock\nalready uses. card/block is cardinality \"many\", so assertFact with an explicit\nid captures the old (parent, position) as one undo entry: undo restores it and\nredo re-applies the move, never passing through a state where the block is gone.\n\nThe forward result is unchanged on both client and server (the server already\ncollapsed the retract+assert of one id into a single put), so this only removes\nthe broken intermediate undo/redo states.\n\nhttps://claude.ai/code/session_017QZrKm9paHWqRQ1yMsSty2\n\n* Fix block disappearing on undo/redo of outdent and child reparenting\n\nExtends the move fix to the remaining retract-then-assert-same-id sites found by\nauditing every ctx.retractFact call in mutations.ts:\n\n- moveChildren: each child was retracted then re-asserted under the new parent\n  with the same id.\n- outdentBlock: the outdented block's own card/block fact, and each following\n  sibling reparented under it, were retracted then re-asserted with the same id.\n\nEach pair produced two undo entries with an intermediate state where the fact\ndoes not exist, so undoing/redoing an outdent (or any op that reparents children)\nblanked the affected block(s) before restoring them. Drop the redundant retracts\nand let the single assertFact overwrite each fact in place (reusing its id),\ncapturing the old (parent, position) as one clean undo entry — the same pattern\nas moveBlock/indent/addLastBlock.\n\noutdentBlock's orphan-prevention early-bail (index === -1) still precedes all\nwrites, so a missing anchor remains a clean no-op. Forward result is unchanged\non both client and server (the server already collapsed retract+assert of one id\ninto a single put).\n\nAll other ctx.retractFact calls are genuine deletions (removeBlock,\ndeleteFootnote, deleteComment, retractAttribute, toggle-off, etc.) and are left\nas-is. editComment intentionally keeps its retract: it asserts a new block/text\nwith no id, where the retract is required to stop assertFact's YJS text-merge\npath from merging two independent docs.\n\nhttps://claude.ai/code/session_017QZrKm9paHWqRQ1yMsSty2\n\n* docs: note mutation-authoring patterns in CLAUDE.md\n\nCapture the undo-correctness rules learned from the recent fixes: move/replace a\nfact with a single id-reusing assertFact (not retract+assert), group\nmulti-mutation actions with withUndoGroup awaiting each mutation, and use\nignoreUndo for derived/follow-up writes.\n\nhttps://claude.ai/code/session_017QZrKm9paHWqRQ1yMsSty2\n\n---------\n\nCo-authored-by: Claude ","tree":"1eac6cc1a32a3711cb9ed92b4efc90dc0a4f8045"},"mode":"0100644","name":"clientMutationContext.ts","size":5174},{"lastCommit":{"author":{"email":"jared@awarm.space","name":"Jared Pereira","when":"2026-06-23T16:58:27.000000-04:00"},"committer":{"email":"jared@awarm.space","name":"Jared Pereira","when":"2026-06-23T17:02:33.000000-04:00"},"hash":"bea4d46aa2ec54bdac89e34a7c6b77c83706df39","message":"Add heading-level folding for document sections\n\nHeadings now own the blocks that follow them until the next heading of\nequal-or-higher level (Obsidian-style sections), derived as a headingPath\non each block alongside the existing list folding. Adds a disclosure\ntriangle, fold/unfold-all shortcuts, and moves that carry a folded\nsection as a unit (moveBlocks mutation) while keeping new/moved blocks\nvisible.\n\nCo-Authored-By: Claude Opus 4.8 (1M context) \n","tree":"dfa85ec8ceb75b0ed722375cce79c41471d98214"},"mode":"0100644","name":"getBlocks.ts","size":8387},{"lastCommit":{"author":{"email":"jared@awarm.space","name":"Jared Pereira","when":"2026-07-18T12:33:41.000000-07:00"},"committer":{"email":"noreply@github.com","name":"GitHub","when":"2026-07-18T15:33:41.000000-04:00"},"hash":"b108a055fd1f8eb74b4e6a27b829d895cce97740","message":"Cut Supabase + Vercel spend without changing user-facing behavior (#355)\n\n* Suppress yjs realtime broadcasts when no other peer is subscribed\n\nA solo editor streams doc updates every 50ms and cursor awareness every\n80ms into an empty channel — every message billed by supabase, delivered\nto nobody. Solo sessions are the common case, so this was the bulk of\n~7.9M billed realtime messages/month.\n\nTrack channel membership via presence and drop doc-update/awareness\nbroadcasts while no other peer is present. Late joiners converge through\nreplicache exactly as they already did for updates sent before they\nsubscribed. Recently received broadcasts also count as peer presence so\npre-presence clients keep working across the deploy boundary.\n\nCo-Authored-By: Claude Fable 5 \nClaude-Session: https://claude.ai/code/session_018MbcWeXrpeshUJLFDJKzek\n\n* Cut database and storage egress in the hot read paths\n\nUncached supabase egress ran 839GB/mo against 53GB cached. Three\ncontributors addressed here:\n\n- Replicache pulls return a full document snapshot every time, and every\n  open tab polled one per minute. Stretch the idle poll to 5 minutes and\n  pull explicitly when the poke channel resubscribes after a drop, so\n  convergence still happens immediately after a missed-poke disconnect.\n  Worst case for a poke lost on a live socket is now 5 minutes of\n  staleness for a passive viewer instead of 1.\n- The bsky feed skeleton endpoints selected entire document rows\n  (including full content jsonb, 25 at a time, on unbounded external\n  traffic) to read only a post ref. Project just uri/sort_date/refs.\n- storage-js expects cacheControl in seconds; passing a full header\n  string stored a malformed Cache-Control (max-age=public, ...) on every\n  uploaded image, defeating CDN caching of user assets.\n\nCo-Authored-By: Claude Fable 5 \nClaude-Session: https://claude.ai/code/session_018MbcWeXrpeshUJLFDJKzek\n\n* Resize images with sharp instead of Supabase image transforms\n\nSupabase bills $5 per 1,000 origin images transformed per month\n(19,275 last cycle). Replace all three transform pipelines with sharp,\nwhich the repo already uses for cover thumbs:\n\n- atproto_images: resize the cached PDS blob in-function rather than\n  via storage.download({transform}); also drops the per-request list()\n  existence probe and reuses fetched bytes instead of re-downloading.\n- link_previews: store the screenshot already downscaled to its 240x208\n  display size and serve the object directly, transform-free.\n- next/image loader: point at a new /api/resized_images sharp proxy\n  (originals come off the storage CDN; responses are edge-cached for a\n  year) instead of the render/image endpoint.\n\nOutput dimensions and no-upscale behavior match imgproxy's; EXIF\norientation is applied explicitly via rotate().\n\nCo-Authored-By: Claude Fable 5 \nClaude-Session: https://claude.ai/code/session_018MbcWeXrpeshUJLFDJKzek\n\n* Serve image bytes from Supabase's CDN via redirect, not through Vercel\n\nThe Vercel invoice shows fast data transfer already ~500GB past the\nincluded 1TB, so every image byte streamed through these routes bills at\n$0.15/GB — worse than the Supabase egress the sharp rewrite was saving,\nand cover/doc images are pure byte volume. Store each resized variant\n(and the cached original) in storage once and 302 to its public URL:\nbytes then leave via Supabase's CDN as cached egress ($0.03/GB with free\nheadroom) and Vercel serves only a year-long-edge-cached redirect.\n\nensureOriginalCached also repairs legacy objects stored with the\nmalformed 'max-age=public, ...' Cache-Control by re-uploading them once\nwith a well-formed value, since redirecting to an uncacheable object\nwould forfeit the CDN saving. Streaming from the PDS remains the last\nresort when nothing could be cached.\n\nTradeoff: one extra redirect hop per image on a cold browser cache.\n\nCo-Authored-By: Claude Fable 5 \nClaude-Session: https://claude.ai/code/session_018MbcWeXrpeshUJLFDJKzek\n\n* Cut Vercel Fluid compute burned on screenshots, favicons, and push logging\n\nFluid bills provisioned memory for full instance wall time, so the\nmulti-second remote-browser renders behind og-images were the main\nburn: every unfurl bot re-fetch after the 60s revalidate paid a fresh\nrender. OG cards are effectively immutable post-publish — raise\nrevalidate to a day and let the CDN serve them (stale by at most a day\nafter an edit). The share-card screenshot route had no caching at all;\nedge-cache it per url.\n\nAlso: the per-leaflet favicon was force-dynamic + no-cache, running two\nDB queries and a satori render on nearly every page load — cache it,\npicking up theme edits within the hour. Gate the per-push performance\nlog summary behind LOG_PUSH_PERFORMANCE. Stop runtime-cache writes for\narbitrary user-pasted URLs in resolveBlueskyPostUrl (near-zero hit\nrate, one billed cache write per paste).\n\nCo-Authored-By: Claude Fable 5 \nClaude-Session: https://claude.ai/code/session_018MbcWeXrpeshUJLFDJKzek\n\n* Reduce Vercel request volume and reader-page bundle weight\n\n- Serve the hot feed from a CDN-cacheable GET route. It's identical for\n  every user and already 5 minutes stale from redis, but each reader's\n  load paid an uncacheable POST invocation returning 50 full posts.\n  Worst-case staleness goes from ~5 to ~10 minutes.\n- Widen replicache pushDelay 500ms -> 1s: halves push invocations during\n  sustained typing. Live co-editing latency rides the yjs channel;\n  replicache-only changes (e.g. block moves) reach collaborators ~0.5s\n  later, and unsent mutations survive tab close in local storage.\n- Load the comment editor and bsky share composer on interaction instead\n  of bundling them: both drag prosemirror into every public post page\n  for readers who never comment or share.\n\nCo-Authored-By: Claude Fable 5 \nClaude-Session: https://claude.ai/code/session_018MbcWeXrpeshUJLFDJKzek\n\n* Drop sub-1%-impact interventions\n\nRanking the branch's changes by predicted monthly savings, three fall\nunder 1% of the total (~$310/mo) because their frequency drivers are\nhuman actions, not machine traffic: quote-screenshot caching\n(share-modal opens), the push perf-log gate (sub-ms CPU per push), and\nthe no-store paste-URL resolution (a runtime-cache write per paste).\nRevert them to keep the diff to changes that pay for their risk.\n\nCo-Authored-By: Claude Fable 5 \nClaude-Session: https://claude.ai/code/session_018MbcWeXrpeshUJLFDJKzek\n\n* Answer conditional feed requests with 304s\n\nThe rss route gets polled harder than any page on the site (156K\nrequests in the last transfer window), and transfer out is billed even\non CDN cache hits, so every poll paid for the same unchanged bytes.\nFeed readers nearly all send If-None-Match; derive an ETag from the\nbody — generateFeed already pins the updated stamp so an unchanged feed\nis byte-stable — and return an empty 304 on match.\n\nCo-Authored-By: Claude Fable 5 \nClaude-Session: https://claude.ai/code/session_018MbcWeXrpeshUJLFDJKzek\n\n* Snap image-resize dimensions to a fixed width ladder (#358)\n\nThe resized_images and atproto_images routes accept caller-controlled\ndimensions, and every novel value mints a permanent stored variant on\ncache miss — an unauthenticated caller could mint ~2000 objects per\nimage (~4M per cid for w×h) just by walking the width space. Snap\nrequests to a 4-entry ladder shared from supabase/imageSizes.js so the\nkeyspace per image is bounded: 360/800 are the app's own\ncover-thumbnail widths, 1200 covers document-body images at retina\ndensity, 2000 is the existing cap. Snapping up (never rejecting) keeps\nURLs already living in edge and browser caches working across the\ndeploy, and off-ladder requests collapse onto the same stored objects\nbecause the snapped value keys both the sharp resize and the variant\npath. atproto_images height is snapped too rather than dropped, keeping\nold cached URLs valid while bounding the w×h grid.\n\nAlign the clients so real traffic is on-ladder before it reaches the\nedge cache: the next/image loader snaps in the browser, and\nimages.deviceSizes is set to the ladder (imageSizes emptied — its\nentries must sit below the smallest deviceSize and every icon-sized use\nalready collapses to 360 via the loader) so srcset candidates come out\nof the same four widths instead of Next's 16-entry default. The share\ncomposer's 1000px cover prefetch moves to the 1200 rung it would be\nsnapped to anyway.\n\n\nClaude-Session: https://claude.ai/code/session_01MQBgS6Lb1BrTgg2pu1Bfm4\n\nCo-authored-by: Claude \n\n---------\n\nCo-authored-by: Claude ","tree":"38933109e777b0b845a3dec50d079a863f2a12b1"},"mode":"0100644","name":"index.tsx","size":9184},{"lastCommit":{"author":{"email":"celine@hyperlink.academy","name":"celine","when":"2026-06-27T22:17:10.000000-05:00"},"committer":{"email":"noreply@github.com","name":"GitHub","when":"2026-06-27T23:17:10.000000-04:00"},"hash":"a1178bad47e80a0c85100316f496d402b6f2b14a","message":"Feature/image gallery block (#325)\n\n* init\n\n* adjustments to the gallery styling, adding alt text\n\n* styling image light box  a little\n\n* some simplication, some view mode optimizations\n\n* added lexicon and published view\n\n* add a cover image from publish\n\n* updates to carousel style, styled the sent state pull in published list\n\n* final tweaks\n\n* last minute sweep","tree":"eab1be36ca9f78208242015c2fc47a19201f1ceb"},"mode":"0100644","name":"mutations.ts","size":37345},{"lastCommit":{"author":{"email":"jared@awarm.space","name":"Jared Pereira","when":"2026-06-04T14:04:36.000000-04:00"},"committer":{"email":"jared@awarm.space","name":"Jared Pereira","when":"2026-06-04T14:04:36.000000-04:00"},"hash":"0522c103d81d2fc9c2ed3a03aa21a6debe945371","message":"clean up unused code and exports\n","tree":"d5196a0277ea5d1e450b599ad0acae6177cf4edd"},"mode":"0100644","name":"useSubscribe.ts","size":2511},{"lastCommit":{"author":{"email":"jared@awarm.space","name":"Jared Pereira","when":"2025-08-25T18:40:27.000000-04:00"},"committer":{"email":"jared@awarm.space","name":"Jared Pereira","when":"2025-08-25T18:40:27.000000-04:00"},"hash":"19377c5eec00f6feac306f4191a5bd34e4fa43f5","message":"fix type of getClientGroup\n","tree":"cbebf39c9d1ab1f406192373021093e57126184b"},"mode":"0100644","name":"utils.ts","size":2852}],"lastCommit":{"author":{"email":"jared@awarm.space","name":"Jared Pereira","when":"2026-07-20T15:27:19.000000-04:00"},"committer":{"email":"jared@awarm.space","name":"Jared Pereira","when":"2026-07-20T15:30:49.000000-04:00"},"hash":"3c851dd284723f6e05f12fbd3ba26df27e2d4c90","message":"cached mutation context handle updates as well as del/write\n","tree":"e2ae6db8d6f0fef6a99e78f5d7f817ffc5ec75ab"}}