Skip to content

Add opt-in native replies with confirmed-send safeguards - #160

Merged
nixfred merged 2 commits into
nixfred:mainfrom
marcus:upstream-replies
Nov 14, 2026
Merged

nixfred merged 2 commits into
nixfred:mainfrom
marcus:upstream-replies

Conversation

@marcus

@marcus marcus commented Nov 10, 2026 •

Copy link
Copy Markdown
Contributor

replies=on enables native Reply for supported direct-message text and single pictures; it is off by default. Quote and reply remains available. The dispatcher, installer and Linux shim explicitly gate the new imsg-reply tool.

Before Return, the tool refuses a Mac draft and requires matching recipient window title, reply target text/time, front app, reply-field identity and exact focused-field contents. Targets with later identical labels at the same HH:MM are refused before opening Messages. Text travels on stdin and is filled through AX without the clipboard. After one Return, chat.db must confirm the outgoing thread. Unconfirmed sends, signal/crash exits and transport/software failures say to check Messages and never automatically hand the words back. Local shim refusal (78) and a distinct never-started sentinel safely restore words; Mac EX_SOFTWARE (70) remains uncertain. BarWidget owns reply jobs and result context across view closure.

Based on current upstream main (69ca3ab), including merged #159 reactions and #163 audio. Preserves upstream theme, controls and manifest; no upstream file deletions. Uses main's shared NativeAX, picker routing, cleanup helpers and permission fixes.

Validation on 6634bdf: 894 Bun tests and every Python/Qt step from CI pass; shell syntax, shellcheck and pyflakes on imsg-react, imsg-reply and blip-dispatch pass. MessageActions and TapbackActions rebuild byte-identically; SendState is regenerated. Synthetic tests cover the fourth-round review's final target/front/focus/placeholder checks, focused-field text, label time, missing Reply transcript, script exceptions, confirmation text and picture-with-balloon refusal. All 11 targeted guard mutations fail tests. No real conversations opened or messages sent.

The recipient-title shape and live macOS behavior still need the maintainer's planned validation. Unsupported title shapes fail closed. Details and limitations are in docs/replies.md.

@nixfred

nixfred commented Nov 10, 2026

Copy link
Copy Markdown
Owner

Thank you, Marcus. Native replies are the feature people ask about most, and the safeguards are the right ones: text on stdin on both machines and never in argv, set through the accessibility value without touching the clipboard, off by default behind replies=on, groups refused on both sides, a log that holds only codes and seconds, and a confirm step against chat.db. An independent review found two blocking problems in the part that decides whether a reply may be re-sent, and I checked both against the code myself.

Blocking

  1. After Return, three failure paths report "nothing sent", so the composer gets the words and the reply target back and a second send can duplicate the reply.
    • In imsg-reply (the JXA send), Return is pressed with se.keyCode(36), then if the field still holds text 0.8 s later the script clears it and returns not-sent, which the tool maps to exit 75, "nothing was sent". Messages being slow to clear the field is not the same as nothing sent. A scripted probe with the reply row landing in chat.db gave exit 75 with one reply row present, and test_reply.py locks that mapping in.
    • A sqlite3.Error or OSError inside confirm(), after Return, is caught by main() and exits 69 (or 70), which BlipView.qml shows as "not sent, Mac unreachable".
    • If ssh drops after Return, the shim turns 255 into 69, the same branch.
      The fix is one rule: from the moment Return is pressed, every outcome that is not confirmed exits 76 (wrap confirm, and treat not-sent after Return as 76). On the Linux side, a reply job should hand the words back only for codes that are provably from before Return, and 69, 255 and 70 should say "check Messages".
  2. The recipient is never verified before Return. select_chat counts it a success when open exits 0 and a window exists, and the bubble scan matches text, HH:MM and side. The only identity check is the chat.db lookup after sending. imsg-read itself records (Astra C#4) that the selection can fail to move. If it stays on another conversation (an alert sheet, or Messages slower than the fixed 1.5 s sleep), a look-alike bubble there ("ok" at 14:05) gets the reply: the wrong person, exit 76. Please check the conversation's identity before Return (the window or header title, or several of the newest transcript labels against the target's rows). How often this can happen cannot be measured without the Mac, which is why it should fail closed.

Should fix

  1. The reply runs in BlipView's sendProc, not BarWidget, although BarWidget.qml records that a long run owned by the view dies when the view closes. A run is 11 to 13 s plus up to 15 s of confirm. Closing the popout loses the result, so the user retypes and duplicates, or the Mac is left with a filled reply field that the next run Escapes into the main composer as a real draft.
  2. Exit 76 never marks the bubble: it stays "Sending..." for the 2 minute pending window and then vanishes silently, and the note shows only if the user is still on that thread.
  3. resolve_picture counts every text-less attachment as an "Includes picture" look-alike (video, file, sticker, several attachments). A later video in the same minute gives ordinal 1, so the scan aims at an older picture. Use the target's own test (one image, not a sticker).
  4. Safeguards with no failing test: the group check in resolve_picture, find_reply's is_from_me = 1 (an incoming reply with the same text would confirm "sent"), the date >= since bound (an older identical reply would confirm), the if not a.yes dry-run gate, the GUID regex, the code === 76 branch in BlipView, and the JXA draft and viewNames checks. The new reply_to.attachment and orig_part code in imsg has no Python test.

Smaller notes: ui.test.ts line 128 was weakened to if (composeField.text === "") {; failPending re-adds the bubble without its reply context; the reply-context auto-fetch skips the bytes-known source gate (the transfer cap still applies); bodyIs now also matches a message that literally reads hi, Sam reacted with joy; and two doc lines overstate: "The reply view must name the target" (it checks text and HH:MM only) and the exit table's "69 = nothing sent" (false after Return).

What I will do. Once 1 to 3 are in, I will run it on the real macOS 27.2 gateway, on the self-thread only, because how long Messages takes to clear the field and whether open imessage: always moves the selection can only be seen on a Mac. This branch and #159 both touch imsg-react, so #159 goes first and this one rebases over it.


Generated by Claude Code

nixfred added a commit that referenced this pull request Nov 10, 2026
…ll 4x faster (CHANGELOG, CHANGE.log, roster, weekly notes)

WHAT: CHANGELOG entries for #158 (Delivered until Read) and #161 (composer emoji picker) with the real end-to-end numbers for the poll speed-up (shallow 0.77 s to 0.2 s, deep 2.1 s to 1.0 s on gus); CHANGE.log entries for 401b99d, #158 and #161; Marcus Vorwaller added to the no-X table in CONTRIBUTORS-X.md (nothing on his GitHub profile or social accounts, his site marcus.pictures lists none, checked 10 Nov); WEEKLY-NOTES W46 section for 10 Nov with Marcus's own words about the rebases and the reasons #159 #160 #162 #163 were sent back.
WHY: the PR-run checklist records every merge the day it lands, with the contributor's real name and a verified X handle or a visible "none found" plus asking Fred the same day.
CONTEXT: independent review of #159 #160 #162 #163 (read-only, nothing sent to the Mac) found: #162 ffmpeg guessed the format from sender bytes (reproduced: a two-line ffconcat file delivered as a video rendered a different cached picture), #163 Qt plays bytes the safety check never saw and playback stops on any list rebuild, #160 three failure paths after Return report "nothing sent" and the recipient is never checked before Return (both confirmed against the code), #159 the shipped macOS 27 path has no test. Comments with the specifics are on each PR. Deployed 10 Nov to the gateway Mac (read_state.py and imsg, backups in ~/.blip/backups-20261110), gus and vic, all healthy, no log errors.
TAGS: docs, changelog, change-log, weekly-notes, roster, marcus, pr-run, review

LR-T: pr-run, contributor-pr, bookkeeping
LR-D: blip, docs
LR-K: pr-158-161-merged, pr-159-160-162-163-sent-back, marcus-no-x
LR-R: 401b99d, ea64c98, 0d302b1

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BXN1v67b9f9VHGe3FhaQuD

marcus commented Nov 10, 2026

Copy link
Copy Markdown
Contributor Author

Addressed the review and rebased onto current main.

  • Slow field clearing after Return is unconfirmed, never nothing sent; confirmation and send-call exceptions preserve uncertainty. Cleanup cannot overwrite the outcome. Linux transport/software failures (69/70/255) and 76 say to check Messages and do not hand the words back.
  • Immediately before Return, Messages must have no sheet and its window title must exactly match the target address/contact name; the reply transcript must also match the target text/time. Focus, identity and transcript are rechecked after filling. Unsupported title shapes fail closed and need your gateway validation.
  • Native reply jobs, their pending context and failure results now live in BarWidget, surviving view closure. Removed the obsolete view-owned reply route. Unconfirmed results remain visibly marked instead of silently expiring; host changes invalidate stale thread loads.
  • Picture ordinals/self echoes use the same one-image/non-sticker predicate as the target. Reply-context auto-fetch now observes the known-source-size gate. Reaction-label lookalikes count as twins. Failed context and the stronger draft handback guard are preserved.
  • Added executable JXA tests for draft, wrong recipient, wrong transcript, focus and slow clear; synthetic tests for confirmation exceptions, old/inbound rows, group pictures, other attachment types, dry-run/GUID gates, bridge orig_part/attachment wire data, and the shipping bar queue functions. The bridge wire regression runs in CI.

Validation: 832 Bun tests, 352 synthetic Mac Python tests, and the reply-banner Qt checks passed. No real conversations were opened or replies sent. CI is running. This still needs your planned macOS 27.2 self-thread validation, especially the title shape; no broad compatibility claim is intended.

nixfred added a commit that referenced this pull request Nov 10, 2026
… (CHANGELOG, CHANGE.log, roster, weekly notes, CLAUDE.md)

WHAT: CHANGELOG and CHANGE.log entries for #162 (native stickers) and the reaction-scan speed-up (9b4f8d9) with the real numbers; CONTRIBUTORS-X.md moves Marcus Vorwaller to 3 merged (#158, #161, #162), still no X handle found and asked Fred; WEEKLY-NOTES W46 gets the #162 and speed-up bullets plus the note that #159, #160 and #163 got a second independent review of their rebased heads; CLAUDE.md's measured-latency paragraph gains the lesson from the reaction scan (the 90-day margin is real data: 3 of 28,305 reactions are stamped before their target, the worst by 0.83 days; an ORDER BY in SQL brings the full scan back; the date index reorders reactions unless sorted by ROWID; do not probe the guid index by shape because _strip_assoc is lenient on purpose).
WHY: the PR-run checklist records every merge the day it lands, and the next agent that "optimises" enrich() must find the traps written down instead of rediscovering them.
CONTEXT: Fred approved the #162 merge and the speed-up from the literal diffs on 10 Nov. Merge commit aee3394, speed-up 9b4f8d9.
TAGS: docs, changelog, change-log, weekly-notes, roster, claude-md, pr-run, stickers, latency

LR-T: pr-run, bookkeeping, latency
LR-D: blip, docs
LR-K: pr-162-merged, tapback-window-bound, marcus-no-x
LR-R: aee3394, 9b4f8d9

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BXN1v67b9f9VHGe3FhaQuD
@nixfred

nixfred commented Nov 10, 2026

Copy link
Copy Markdown
Owner

Second-round review of ed13570, thank you for the fast turnaround. An independent reviewer went through it read-only (nothing sent to a Mac); I re-ran the main blocker and checked the others in the code.

Fixed and confirmed

  • The Mac side of my first blocker: every failure after Return now exits 0 or 76 (the reviewer drove the real main() with a scripted osascript and built the path table), and nothing re-sends by itself.
  • The conversation is verified before Return (window title against the chat id or contact name, no sheet open, the reply view shows the target's text and time), rechecked immediately before Return; a switched conversation, a different chat on screen or an alert all refuse.
  • Message text never rides argv or script source; replies=on is parsed, not sourced, and off by default; the ordinary send path is unchanged.

Blocker 1: the PR breaks tapbacks that work on main today. bridge/mac/imsg-react:254 uses REACTED, which is never defined on this branch (it looks copied from #159 without its definition; I checked: that is its only occurrence). When a message has a later message from the same side at the same HH:MM that starts with ", ", both lookups crash. Repro with the reviewer's script on your head: target "ok", later "ok, see you then" gives NameError: name 'REACTED' is not defined from both the tapback path and the reply path; the plain later "ok" works. Main resolves the same data. No test covers that branch and the CI compile step cannot catch an undefined name; python3 -m pyflakes on imsg-react would.

Blocker 2: my first blocker is only half fixed on the Linux side. BarWidget.qml finishReply (~152) marks every non-zero exit as failed, and the bubble then shows the red "Not Delivered". replyFailure in send-state.ts (line 29) lists only 69, 70, 76 and 255 as "Check Messages before sending again"; every other code, including a process killed by a signal, gets "Reply was not sent", which may be false. The words are correctly not handed back, but the bubble still invites a retype and a duplicate.

Should-fix

  1. In-flight replies vanish after 2 minutes. withPendingSends (thread.ts ~461) drops pending entries older than PENDING_MAX_AGE_MS unless they already failed, and reconcileReplies removes the reply while its job is still running; a later 76 then marks nothing. The reviewer's repro passes against the shipping functions. Runs can exceed 2 minutes: the Mac UI lock waits up to 150 s and replies queue one at a time.
  2. The rebase over Support tapbacks on older messages and exact emoji on macOS 27 #159 needs more than resolving markers. The branch carries a stale copy of Support tapbacks on older messages and exact emoji on macOS 27 #159's code (NativeAX including the synthetic-click path, close_picker, clear_stale_picker, the emoji JXA helpers) and adds a check that the tapback picker button is enabled, which changes today's tapbacks. A trial merge on top of Support tapbacks on older messages and exact emoji on macOS 27 #159 conflicts in ci.yml, TapbackActions.mjs, tapback-actions.ts and four imsg-react hunks, and git merges the duplicate class and functions without markers: in a marker-only resolution Add opt-in native replies with confirmed-send safeguards #160's stale NativeAX shadowed Support tapbacks on older messages and exact emoji on macOS 27 #159's and the event-access permission check disappeared. Running pyflakes on the merged file flags all three duplicates, so it is a cheap gate. Please rebase onto Support tapbacks on older messages and exact emoji on macOS 27 #159 once it lands (order is Play voice messages inline with real waveforms and seeking #163, Support tapbacks on older messages and exact emoji on macOS 27 #159, Add opt-in native replies with confirmed-send safeguards #160) rather than carrying a copy.
  3. The new CI step runs zero tests. bridge/mac/test_reply_enrich.py has no unittest.main() (zero matches), so python3 bridge/mac/test_reply_enrich.py exits 0 without running anything; with the attachment-part logic deliberately broken the CI command still passed.
  4. The guard that decides whether the user's words come back is untested. These mutations all passed both suites: adding 76 to the hand-back list; handing the words back on every exit code; dropping the sheet check; dropping the recipient recheck after filling the field; no longer limiting the transcript check to the reply view. 8 of 15 mutations were caught.

Still unproven (needs the gateway Mac, self-thread only, with Fred's explicit word): what Messages' window title looks like on 27.2 and whether a sheet shows while the reply view is open, whether open imessage: always moves the selection, how long Messages takes to clear the reply field, how slow the reply-view check is on long conversations, and whether replying marks the conversation read.

bun test 841 pass, the CI Python suites and the Qt suites pass on your head, and the rebuilt QML modules match their sources.

nixfred added a commit that referenced this pull request Nov 11, 2026
WHAT: the same-turn bookkeeping for the #159 merge (6a4fbce, 56bc294).
  - CHANGELOG.md: a proper Unreleased entry; the one-line note Marcus's rebase
    left above the "## Unreleased" heading is removed (it was orphaned there).
  - CHANGE.log: #159 entry with the live 27.2 proof and what Larry added.
  - CONTRIBUTORS-X.md: Marcus 5 merged (#158 #161 #162 #163 #159), #160 under
    review, X handle still none found (rechecked 2026-11-11, asked Fred).
  - WEEKLY-NOTES.md: W46 Wed 11 Nov, #159 bullet and the two Blip-side findings.
NOT staged on purpose: bridge/mac/findmy_city.py and test_findmy_city.py are
  someone else's uncommitted work in this tree.

LR-T: blip, pr-159, changelog, contributors, weekly-notes
LR-D: blip, docs
LR-K: pr-run-bookkeeping
LR-R: 6a4fbce, 56bc294

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

nixfred commented Nov 11, 2026 •

Copy link
Copy Markdown
Owner

Third-round check of d0fc54f. This time I ran the checks myself on your head, no independent reviewer yet: that full read waits for the rebase, because the code that ships is what is left after it. #159 and #163 are both merged.

Fixed and confirmed (I ran each)

  • Blocker 1 is gone: REACTED is defined on this head, pyflakes is clean on imsg-react, imsg-reply, imsg and blip-dispatch.
  • Blocker 2 is gone: replyFailure now says "Check Messages before sending again" for every exit code that does not prove nothing was sent, and the words come back only for [1, 64, 65, 75, 77] (replyCanRestore). Adding 76 to that list, or handing the words back on every code, each fail 2 tests.
  • The 2-minute drop is fixed: an in-flight reply survives PENDING_MAX_AGE_MS in withPendingSends, and reconcileReplies keeps it. Removing either fails a test.
  • Zero-test CI is fixed: every Python test file now has a main guard (test_reply_enrich.py runs 1 test, test_reply_jxa.py 1, test_reply.py 29).
  • bun test 875 pass, the three Python suites and test-reply-ui.py (3 tests) pass, CI is green, and MessageActions.mjs, SendState.mjs and TapbackActions.mjs rebuild byte-identical.

Still open

  1. The rebase. The branch is on 94fb8c7. Trial-merging it onto current main (369f722) conflicts in four files: .github/workflows/ci.yml, TapbackActions.mjs, tapback-actions.ts and bridge/mac/imsg-react. This is the stale copy of Support tapbacks on older messages and exact emoji on macOS 27 #159 I flagged: after the rebase, imsg-react should contain only Add opt-in native replies with confirmed-send safeguards #160's own hunks, with Support tapbacks on older messages and exact emoji on macOS 27 #159's NativeAX, picker_route, close_picker and clear_stale_picker taken from main as they are, and TapbackActions.mjs regenerated from the merged tapback-actions.ts rather than conflict-resolved by hand. When you push it, run python3 -m pyflakes bridge/mac/imsg-react once: git can merge a duplicate class without markers and a stale copy then shadows the new one. Main also has two fixes in imsg-react's permission probe (33aab7b and ef9a1ab) that you will pick up in the rebase.
  2. Guards with no test (mutation check I ran this round: 12 mutations, 8 caught, 4 not). These changes left every suite green:
    • BlipView.qml ~2004: dropping composeField.text === "" from SendState.replyCanRestore(code) && composeField.text === "". A failed reply could then overwrite a draft you have started since.
    • imsg-reply send(): dropping the "reply field already holds text" refusal (if (i.length > 0) return { ok: false, error: "draft" }), so Blip would overwrite a half-typed draft in the reply view and send it. This is the one I would test first.
    • imsg-reply send(): dropping the identity(q) check before the field is filled, and dropping viewNames(q) before the field is filled. The re-check after filling still catches both, so these are weaker; one test that the early refusal happens before any value is set would cover both.
    • Caught: the sheet check, the recipient re-check after filling, and the transcript check being limited to the reply view.

Still unproven, needs the real 27.2 gateway, self-thread only, with Fred's explicit word (I send nothing without it): what Messages' window title looks like on 27.2 and whether a sheet shows while the reply view is open, whether open imessage: always moves the selection, how long the reply field takes to clear, how slow the reply-view check is on a long conversation, and whether replying marks the conversation read.

Thank you for the quick turnaround on the blockers. After the rebase I will do the full independent review and the live checks.


Generated by Claude Code

Gate native replies behind replies=on and confirm one Return through chat.db. Keep uncertain sends visible and never restore text unless the exit code proves a refusal; preserve newer drafts and bar-owned reply jobs.

Rebase onto current upstream reactions and audio, keeping shared AX helpers and permission fixes intact. Regenerate the QML module and execute draft, recipient and target guards before any write; mutations of all four review gaps fail the tests. Preserve upstream UI, manifest, stdin transport and synthetic-only validation.
@nixfred

nixfred commented Nov 13, 2026

Copy link
Copy Markdown
Owner

Fourth-round review of bfaa399, thank you for the rebase onto #159 and #163. This is the full read I promised, by an independent reviewer (read-only, nothing sent to any Mac), plus my own mechanical checks. There is no blocker and the Mac side is in good shape. I will merge once the four items below are fixed and the live checks are done.

Fixed since last round and confirmed

  • The reviewer built the path table for imsg-reply from start to Return: every failure after Return exits 0 or 76, nothing resends by itself, and no realistic failure sends to the wrong person or invites a duplicate on the Mac side.
  • Ran by me: trial merge on main 69ca3ab is clean; pyflakes is clean on imsg-reply, imsg-react, imsg and blip-dispatch; gitleaks is clean; every new Python test file has a main guard; on the gateway Mac's own Python 3.9.6, test_reply.py (29), test_reply_enrich.py and test_reply_jxa.py pass. Ran by the reviewer: bun test 890 pass; each new Python test file fails when its code is broken; the reply UI test passes; the four generated .mjs files rebuild byte-identical; the Support tapbacks on older messages and exact emoji on macOS 27 #159 helpers are not duplicated.

To fix before merge

  1. A shim killed by a signal can hand the words back. BarWidget.qml ~165: onExited: function(code, status) { root.finishReply(code) } ignores status (I read that line). The reviewer ran Qt's QProcess: a child killed by SIGHUP reports exitCode 1 with CrashExit, and 1 is in replyCanRestore's [1, 64, 65, 75, 77], so the text returns as "not sent" after a send that may have happened. Not run against Quickshell's own Process (I expect the same signature, as reactProc relies on it). Fix: treat any non-normal exit status as uncertain, whatever the code.
  2. Twins are aimed by count, not refused. Two identical "ok" from me in one minute, or the same picture at the same HH:MM on another day, get an ordinal (the reviewer ran both). viewNames() only checks that some label has the target's text and time, so it cannot tell twins apart or check the side. The reply goes to the right person and confirm then reports "unconfirmed", but it can thread under the wrong bubble. Simplest fix: refuse in imsg-reply when ordinal > 0, or say plainly in docs/replies.md that twins are aimed by count.
  3. The last check before Return does not tie the focused field to the one that was filled. With focus moved to an empty field after the fill (shipped JXA, ok returned), Return is pressed there, nothing is sent, and the reply text is left behind as a draft in Messages on the Mac after Escape. Fix: also require the focused field's length (or value) to equal the body in the line-139 recheck.
  4. Untested guards. 21 of 58 mutants survived. The ones that matter: the recheck right before Return (viewNames, front(), focus id and placeholder), the time compare in viewNames, "no Reply transcript means pass", an exception from the send script turned into a "nothing sent" Stop, find_reply ignoring the text, and the picture-with-balloon refusal. The full list with the exact edits is in the reviewer's report: file:///tmp/claude-1000/-home-pi-nas-Projects-blip/cd5fefb7-92e7-49de-8b71-5a0bd0e659d3/scratchpad/review-160-r4.md

Small UX item (safe): some provably-before-the-Mac failures say "Delivery unconfirmed" and keep the words: shim exit 78, a process that never starts (finishReply(70)), Mac-side 69. Adding 78 to the hand-back list is safe (the shim exits before ssh); do not add 70, it is also EX_SOFTWARE from the Mac, so give a never-started process its own sentinel.

Live checks (mine, with Fred's word, self-thread only; nothing runs until he says so): the window title shape for a contact, a bare number, an email and the self-thread, and whether a sheet shows while the reply view is open; what Escape does with text in the reply field; whether setting the field value then Return sends exactly that text (line breaks, emoji) and how long the field takes to clear; picture discovery and labels; how slow the checks are on a long conversation against the 45 s timeout; whether replying marks the conversation read.


Generated by Claude Code

Treat signalled helpers as uncertain, reserve a distinct local not-started result, and refuse later twins before touching Messages. Require the focused reply field to retain the exact body before Return. Add synthetic regression and mutation coverage for the reviewed send guards; preserve the no-automatic-resend invariant.
nixfred added a commit that referenced this pull request Nov 14, 2026
WHAT: entries for Marcus Vorwaller's opt-in native replies (#160, merge e7ac796),
  his roster count (6 merged, no X account, credit by GitHub handle), W46 Sat.
WHY: same-turn bookkeeping for a merge; four review rounds, all fixed.
NOT DONE: live send on the 27.2 gateway, which needs Fred's explicit word.

LR-T: blip, pr-160, changelog, contributors, weekly-notes
LR-D: blip, docs
LR-K: pr-160-merged-bookkeeping

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BXN1v67b9f9VHGe3FhaQuD
@nixfred
nixfred merged commit e7ac796 into nixfred:main Nov 14, 2026
1 check passed
@nixfred

nixfred commented Nov 14, 2026

Copy link
Copy Markdown
Owner

Merged as e7ac796. Thank you, Marcus: four rounds in five days, and every point answered in code within hours. Native replies are a big addition and you kept the safety model intact the whole way.

What I checked on 6634bdf before merging: all four round-4 items are fixed (a signal-killed helper now counts as unconfirmed, twins are refused before Messages is touched, the focused field must hold the exact text before Return, the never-started helper has its own sentinel). I broke each new guard on purpose, ten ways, and every break failed a test. CI is green, bun test 894 pass, the reply Python suites pass on gus and on the gateway Mac's own Python 3.9.6, pyflakes and gitleaks are clean, and the generated modules rebuild byte-identical.

It stays off by default (replies=on). One thing is still unproven: the live macOS 27.2 behaviour (the window title shape, and that setting the field value then Return sends exactly that text). That needs one real send in Fred's own thread, which I only do with his explicit word, and I will report here when it is done. Credit in the update posts will read "Marcus Vorwaller (github.com/marcus)", as you asked.


Generated by Claude Code

nixfred added a commit that referenced this pull request Nov 16, 2026
…t label

Jason Alexander's fix for a contact texting on RCS while the chat row still
said iMessage (every send left as iMessage and died with error 22). The send
service now comes from the opened conversation's own rows via threadSendService,
with SMS only ever on phone numbers and short codes, groups never carrying
--service, and an RCS business agent exception.

Conflicts with #160 (BlipView.qml send block, SendState.mjs, send-state.test.ts)
resolved here because pushes to contributor forks are blocked by the Law 11
guard: kept #165's sendServiceFor and #160's Date.now()-based localId, kept both
sides' tests, regenerated SendState.mjs with the command in send-state.ts.

Checked on 3a13a27 merged with main: bun 927 pass, send-service/reply/read/
reaction QtTest scripts pass, generated module identical to the generator,
gitleaks clean. NOT yet proven on a real Android RCS contact through the 27.2
gateway.

LR-T: blip, pr-165, send-service, rcs, merge
LR-D: blip
LR-K: send-service-from-conversation-rows
LR-L: law-11

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BXN1v68N3DWV2UhtuJggZz
nixfred added a commit that referenced this pull request Nov 17, 2026
…d voice messages

WHAT: Unreleased becomes 2.7.0 (2026-11-17) with a short "Before you update"
block on top (re-run blip-setup on the Mac, the optional ffmpeg and Qt6
packages for voice and stickers, the opt-in keys replies=on, tapbacks=on and
shared_city=on, and the two things not yet proven live). manifest.json and the
README version badge go 2.6.2 -> 2.7.0, and the README test badge goes 790 ->
927 passing, the real count. The CHANGELOG heading uses hyphens, not the
em-dashes of the older headings, because Fred's no-em-dash rule covers every
file; the version-agreement test only needs "## 2.7.0" at the line start.
WHY: Fred, 2026-11-17: asked to consider a 2.7.0 release, was shown the
readiness check (200 commits from 22 people and 42 merged PRs since 2.6.2, all
of it already written up under Unreleased, no open PRs), then ordered the
official release. 2.6.2 (26 Sep) was the latest release for seven weeks.
VERIFIED: bun test on gus at c95b556, the tree before this commit: 927 pass,
0 fail. HEAD, origin/main and forgejo/main were all c95b556 and gh listed no
open PRs. The ui.test.ts version-agreement test passes with the three files in
step, and gitleaks found nothing in the diff. The suite on the larry VM shows
one failure only because ffmpeg is not installed there (audio-waveform.test.ts
shells out to it); gus has ffmpeg and is clean.
OPEN: a send to a real RCS contact through the Mac (#165) and a native reply on
a 27.2 gateway (#160) are not proven live; both are said so in the notes.
CONTEXT: staged by path, the three release files only. TRIAGE-2026-10-02.md and
pitch_deck_content.md are untracked and not part of the release.

LR-T: blip, release, 2.7.0, changelog, version, contributors
LR-D: blip, release, github
LR-K: blip-2-7-0-release-2026-11-17, blip-release-after-2-6-2

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BXN1v3LkbABkmYshbwkWb6
nixfred added a commit that referenced this pull request Nov 17, 2026
…ger say the release is open)

WHAT: CHANGE.log gets the 2.7.0 release entry (newest first, feeds X-POSTS.md):
tag v2.7.0 on 9fe5d98, the headline PRs, the 21 outside contributors thanked by
handle, Erik Fillipsveen named, and the two things not proven live (#165 on a
real RCS contact, #160 on a 27.2 gateway). WEEKLY-NOTES.md W47 loses the line
"A release is Fred's call", which stopped being true when Fred ordered it, and
records the release instead, with a note that the post for it names Erik.
WHY: the notes are written the day the work lands, and a notes file that still
says the release is open sends the next run off to ask Fred again.
VERIFIED: docs only, two files staged by path; gitleaks clean on the diff. The
release itself was verified in 9fe5d98: bun test on gus 927 pass 0 fail, origin
and forgejo main and the peeled v2.7.0 tag all at 9fe5d98, the GitHub release
live and marked Latest, raw manifest.json on main says 2.7.0.
OPEN: the 2.7.0 announcement post is not written; Fred posts, Larry drafts.

LR-T: blip, release, 2.7.0, change-log, weekly-notes, w47
LR-D: blip, docs, release
LR-K: blip-2-7-0-release-2026-11-17, blip-release-after-2-6-2

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BXN1v3LkbABkmYshbwkWb6
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.

2 participants