fix: docs-audit bug batch (headless, reading helpers, held tasks, standing, teams, fan-out checks, media, remote) by santoshkumarradha · Pull Request #1604 · Agent-Field/CodeAF

santoshkumarradha · 2026-09-27T06:07:39Z

One stacked branch for the docs-audit bug list (BUGS-MASTER.md, dev 837b2b0). Each commit fixes one issue or one closely related group. Every fix has a named test. For the critical fixes and most others, the reviewer confirmed that the test fails with the fix reverted. Draft: not for merge until the owner has looked.

Fixed

Issue	Fix	Test (fails without the fix)	e2e (real binary, isolated HOME, default model)
#1566	A blank brief is refused where the goal text is resolved, on every road	`TestReadTextRejectsAnAllBlankArgumentList`	PASS: `codeaf do ""` and `do " "` exit 1 with `error: no goal given`; 0 usage rows
#1567	Delegate flags are parsed wherever they sit; an unknown flag is refused; `--` ends flags	`TestParseReadsFlagsAfterTheBrief`, `TestParseRefusesAnUnknownFlagAfterTheBrief`	PASS: flags-after-brief run shows `maxCost=$0.5`, `ceiling_usd:0.5`; `--bogus-flag` exits 1 naming it
#1568	A read hand-off gets the audit read-only belt (durable flag on the quick node); what it cannot do goes back to the conversation	`TestQuickWorkerCarriesReadOnlyBelt`, `TestQuickReadOnlyBeltRefusesWrites`	not run (model-dependent turn shape)
#1549	Run workers carry the one standing section through the common brief composer	`TestBeltWorkerBriefCarriesStandingOrders`	PASS: 2/2 worker briefs carry `Standing orders` and the rule
#1551	Manager gets a fallback handle at registration; `team_start` carries the approval posture; fallback rule words show the cost clause	`TestAnUntitledManagerGetsAHandleWhenRegistered`, `TestTeamStartOfKindTeamMakesASubTeamWhoseManagerReportsUp`, `TestTheStartsCardSaysWhoIsStartedAndWhatItCosts`	PASS (item 1): new manager has `"handle":"manager-dohh"`
#1552	`M` gives the keyboard to the new manager; hints say `M`; key sheet lists `M m p r d u`	`TestStartingAManagerReturnsKeyboardToItsComposer`, `TestTeamsActionKeysAreListedOnTheHelpSheet`	—
#1553	Redial reclaims the keyboard by client id (only the same window); a redial's ssh stderr goes to the diagnostic tail	`TestARedialReclaimsTheKeyboardBeforeTheOldPipeLeaves`, `TestRedialSSHStderrStaysOutOfTheTerminal`	—
#1555	Project orders compare cleaned paths; card quotes the live allowance; task firings read `task:`	`TestProjectStandingOrdersMatchANormalizedWorkspace`, `TestStandingCardReadsTheLiveDailyBudget`, `TestATaskFiringUsesTheTaskActivityWord`, `TestATaskFiringIsDrawnAsATask`	—
#1557	`/drafts` draws its page; an empty `/skill` closes and clears the box (folder refusals kept)	`TestEmptyDraftPageDrawsItsOwnEmptyState`, `TestBareSkillOnAnEmptyShelfDoesNotLeaveAPathInTheBox`	—
#1572	Checker ceiling kept per check task; unfinished checks fail the run and are named	`TestCrewFactoryGivesEachCheckerItsOwnSpendCeiling`, `TestSupervisorDoesNotFinishWhenACheckDoesNotFinish`	not exercised: the planner did not split in two tries
#1576	Team header reads today's decided raise	`TestTeamsHeaderUsesTodaysRaisedCap`	—
#1577, #1583	Wall shows only this window's conversations; a tile lands on its conversation	`TestWallTileFromTeamsPageLandsOnConversation`, `TestSharedEngineWallDoesNotKeepTheSwappedConversationOpen`	—
#1578	Inline renderer keeps `<id>`-style text	`TestInlineAngleBracketPlaceholdersStayInModelProse`	—
#1580	ssh control path must fit with OpenSSH's temp suffix too	`TestSSHControlPathFitsOpenSSHTemporarySuffix`	—
#1581	Team-name answer wakes the paint clock	`TestTeamNameAnswerWakesAnIdlePaintClock`	—
#1585	`team_start` description no longer claims the person is always asked	`TestTeamStartDescriptionMatchesTheConversationPosture`	—
#1586	Images named by sniffed bytes, one helper for both doors	`TestGenerateImageNamesTheFileByItsBytes`, `TestGenerateImageUsesTheSniffedExtensionWhenTheProviderLies`	—
#1587	esc cancels surface-owned browser sign-ins	`TestEscapeCancelsAPendingCodexBrowserSignIn`	—
#1588	Speech bills under its own role; media calls write a call-log pair with cost	`TestSpeakUsageLedgerNamesTheSpeechCall`, `TestSpeechCallIsWrittenToTheCallLogWithItsCost`	—
#1589	Composing list admits audio rows with the music id mark	`TestComposingPickerKeepsAudioRowsMarkedAsMusic`	—
#1591	Naming its own team in `team_post` gets the right rule	`TestSubTeamManagerNamingItsOwnTeamGetsTheRule`	—
#1535	Hosted surface reaches the engine's memory commands; welcome says whether it remembers	`TestMemoryCommandsUseTheHostedStoreWhenTheAgentHasNoLocalBrain`	—

Fixes #1566
Fixes #1567
Fixes #1568
Fixes #1549
Fixes #1551
Fixes #1552
Fixes #1553
Fixes #1555
Fixes #1557
Fixes #1572
Fixes #1576
Fixes #1577
Fixes #1578
Fixes #1580
Fixes #1581
Fixes #1583
Fixes #1585
Fixes #1586
Fixes #1587
Fixes #1588
Fixes #1589
Fixes #1591
Fixes #1535

Partly fixed (issue stays open)

Squash-merge note: some commit messages below say Fixes #1571, Fixes #1554 and Fixes #1574. Delete those lines from the squash message, because these three issues are only partly fixed.

A held (queued) task cannot be stopped and keeps the repo write-lock #1571: a held task owns no folder lock or branch (e2e: git branch -a shows only main while it is queued). It stops at once from its task room (x → stop it): the room reads stopped, and git status is clean. Still open: /stop from the main box says open a running task to stop it and does not offer the queued task. The held row also shows no machine-busy reason, and its room header reads working (that part is A task held as machine busy shows no reason; max_load changes need a restart #1579).
session: tasks stuck in working on restart, stopped status mapped to failed, and reader handoff #1554: item 2 is fixed. A stop reads stopped with no /redo stronger (e2e: crew line task 1 crew · stopped · …), and the router learns nothing from it. Item 3 is fixed: a failed read hand-off retires its helper and leaves no stopped card. Item 1 (a task stuck after a restart) is not fixed. The first attempt made a never-run task read done, which was worse, so commit 8254cb07d takes it back.
Run ends 'branch kept' with nothing blocking; /land says nothing is waiting #1574: .orig backups are now droppings, and kept, conflicted or aborted run branches are listed and landable by /land. Still open: a single task's kept branch is not listed by /land (e2e, twice). branch kept on main is the deliberate "tasks do not merge into main automatically" guard, so its first acceptance line is an owner ruling.
'Run this as a task on <model>' becomes a reading helper; chat misreports it #1569: the helper can no longer edit files in place (read-only belt). The chat's prose quoting a model, time and cost comes from the model, not from code, and is unchanged.
A task typed on Home can land in another project's repository #1466: a conversation whose engine answers for another workspace is now refused visibly before anything is sent. It is never used. The trace found no code path where a Home /task resolves its target differently from a plain message. The road that reaches the wrong engine is not yet found.
cli: garbled usage text, omitted help flags, zero token reporting in do --json, and workspace defaulting #1558: the token count in do --json is fixed. Not changed: engine --status/--stop with no --workspace means the home folder by design (the code comment explains ssh-launched engines), so that is an owner ruling. The usage-text items were not found on this base.
chat: slash command palette omissions, conversation title fallbacks, raw error text, and status hints #1559: the / list now names /land now, /crew cap task, /cache clean now. /compact says nothing to compact yet, and task titles end on a whole phrase. The /status wording change was dropped: it contradicted a stated rule that the page says the posture, not the badge. That needs a ruling.
tasks column: check tasks omitted from roster, landing race drops card, and room input key capture #1556: check rows are no longer deduplicated away by title. The landing-race and room-input items are open.

Needs owner ruling (not changed)

session: interactive chat bypasses daily spending limit #1546, daily limit in chat: the owner named this one as a product decision.
Default model answers in tool markup; recovery on Opus is over half the spend #1570, mark reader on the top tier: the tier is a stated choice backed by a benchmark (the cheap reader left 17–53% of marks unanswered). The spend evidence argues the other way.
Default per-team cap also caps All teams, so the whole org shares one pool #1575, default per-team cap as one pool: the ancestor-pool model is written as design. Only the header grammar (All teams's cap → All teams' cap) is fixed, in the teams commit.
A task held as machine busy shows no reason; max_load changes need a restart #1579, live task.max_load: the code states that ceilings are resolved once per session, so a run's rules do not change under it.
senior-dev ends 'unsubmitted' (exit 2) after a passing verification #1584, senior-dev unsubmitted after a passing check: the documented contract is that verification is not submission.
A shell OPENAI_API_KEY overrides the profile key; its 401s retry as 5xx #1590, profile key vs shell OPENAI_API_KEY: the key order is documented in the manual as env first. The 401 retry half was dropped with it and needs a fix after the ruling.
Run ends 'branch kept' with nothing blocking; /land says nothing is waiting #1574, first half: see above.

Dropped or left to other PRs

tui3: slash command completion inserts bare command prefix into prompt instead of navigating #1548, session: standing execution falsely reports branch isolation while committing to main #1550, A truncated streamed reply is accepted as a complete answer #1467: already have open PRs (tui3: allow slash command completion to execute bare /task, /redo, and /workspace immediately (#1548) #1563, fix(session): standing execution falsely reports branch isolation while committing to main (#1550) #1560, Discard incomplete streaming replies #1497), so they are not duplicated here.
tui3: task proposal countdown swallows keypress and starts declined task on enter #1547: does not reproduce with a unit test of the shared question router on this base. The worker's change (letting a digit pass the quarter-second settle guard on countdown cards) weakened a documented guard, so it was rejected. It needs a tmux replication.
A standing firing that changed nothing is logged as landed #1582 (a no-change firing logged as landed): the change reversed a stated rule ("a sentence counts as something"), so it was rejected. It needs a ruling.
Codex sign-in link runs across the Tasks column #1592, Settings search does not find the media rows #1593: already fixed on this base. The worker's regression tests were not kept.
pool: a chat /task never reaches the Model Pool judge #1527: not attempted (unconfirmed).
Planner-declared checks keep an unfilled placeholder (issue_.go) #1573: dropped after one re-brief. The first attempt guessed filenames from _ stems, which was a band-aid. The second refused any check path that neither exists nor is a declared target, but the full suite showed it refused legitimate checks on packages a part creates (TestASharedCheckDeclaredByEveryPartIsLiftedOffTheirOwnDoors). Commit take back the part-check path refusal removes it. The trace showed no code drops a placeholder: the planner writes issue_.go literally.

How it was checked

The reviewer read every diff. For codeaf do "" runs a paid job with an empty brief #1566, senior-dev silently ignores --max-cost and other flags after the brief #1567, Manager brief taken by a reading helper that fakes the team's work #1568, session: plan-born task worker briefs omit standing project orders #1549, teams: manager conversation initialization, handle fallback, and approval posture inheritance #1551, teams page: keyboard focus lost on manager creation, key hints mismatch, and roster filtering #1552, remote: --host redial loses keyboard focus and dumps ssh stderr over alt-screen TUI #1553, standing: place scoping mismatch for project orders, stale budget on consent card, and activity label #1555, tui3: input-trapping empty/unwired places for /drafts and /skill #1557, Checker spend ceiling is shared by the whole run, and the run still says done #1572, Opening a wall tile from the teams page lands back on the teams page #1577, A new window's wall shows another window's conversation as open here #1583, Angle brackets in model prose are dropped or double-escaped #1578, Team naming card does not paint until the next key #1581, The composing model list in /settings is empty #1589, memory place says 'memory is off' on the default engine road while memory is on #1535, Generated pictures are JPEG bytes saved under a .png name #1586, A pending browser sign-in cannot be cancelled with esc #1587, session: tasks stuck in working on restart, stopped status mapped to failed, and reader handoff #1554 and A held (queued) task cannot be stopped and keeps the repo write-lock #1571, the reviewer reverted the fix, saw the named test fail, and restored it. For After 'Raise to $4' the team header still shows the old cap #1576, Manager tells the person it asked for approval when nobody was asked #1585, Speech spend is missing from usage.jsonl and the call log #1588, Sub-team manager naming its own team in team_post gets a misleading refusal #1591 and --host fails under a long codeaf home: ssh control socket path too long #1580, the worker's before-and-after runs were relied on.
make test-quick passes: build, vet, gofmt, packed manual, change entries, manual gates, laws.
make test PKGS=./... passes except 10 internal/tui3 credit-warning tests. Those fail the same way on this machine on a tree with no internal/tui3 changes (base plus the headless commits), and there is no known-red ledger file on this base.
Under full-machine load, the last internal/session shard runs each showed one different test failing (TestClosingUnderAProgramsRunEndsItInItsStoreFirst, TestTheReflexCallsAreOnTheSessionsBill). Both pass when rerun alone (-count=3), and the first one also fails on the base-plus-headless tree, so they look like load flakes, not regressions.
make build passes. The e2e above ran on bin/codeaf at 2e7a862, with an isolated HOME, the default model and private tmux servers, for about $0.21 in total. Commits after that are the change entry and the restart revert.

🤖 Generated with Claude Code

readText resolved one empty or all-blank argument to an empty goal with no error, so `codeaf do ""` ran a paid job with a goal the planner invented. The blank check now sits where the goal text is resolved, for every road. Fixes #1566 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Go's flag package stops at the first positional word, so the documented `senior-dev "<brief>" --max-cost 0.5` form folded every trailing flag into the brief and ran with the default ceiling. Flags are now read wherever they sit for every delegate program; `--` still ends them. Fixes #1567 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

The run engine's workers reported dollars but not tokens, so `do --json` printed zero tokens beside a non-zero spend. Workers, the supervisor and the summary now carry input and output tokens to the receipt. Part of #1558 (the token item; the other items stay open) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Browser sign-ins started from /connect or the Codex row live on the surface, with no turn for esc to interrupt, so the listener stayed pending. Esc now cancels every surface-owned browser flow through the existing abandon door. Fixes #1587 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

The read hand-off admitted an ordinary quick task with the full belt (bash, commit), guarded only by a prose line, so a manager's whole directive could be taken by a "reading" helper that wrote, committed and merged, and a request for a task on a named model became in-place edits with no branch. The quick node now carries a durable read-only mark and its worker gets the audit read-only belt at the one place quick workers are built; a request the reader cannot serve hands control back to the conversation. A failed hand-off retires a never-started helper instead of cancelling it into a stopped card, and still stops one that did start. Fixes #1568 Part of #1569 (the chat's own prose about model and cost is unchanged) Part of #1554 (item 3) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

TaskGraph.briefLocked appended the project's standing orders to a task node's brief, but a run's workers are briefed from the plan store by BeltWorkerBrief, which never saw that tail, so no plan-born worker was told the rules. The run spec now carries the one rendered standing section and every worker the run seats gets it through the common brief composer. Fixes #1549 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…llowance, task firings say task A project order stored with a trailing separator did not reach its own project and was listed under other projects; the standing card quoted the day allowance the session started with, not the one /budget set (the one the spend rail enforces); and a firing that ran a task was drawn as "said:". Fixes #1555 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

/drafts claimed every key while its page had no overlay height or rows, so it drew nothing until esc. /skill on an empty shelf opened a picker with no choices and left `/skill ` in the box, so later commands became its path. The drafts page is drawn and pointed like every other place, and an empty skill catalog closes the picker and clears the box once both reads answer. Fixes #1557 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…the cards and pages say what is true - A manager made with M had no handle until its first turn titled it, so team_send from the global manager could not reach it. SetManager now gives an untitled manager a unique fallback handle when it is registered. - A sub-team started by team_start did not inherit the starting manager's approval posture; the start entry now carries it and the child gets it before it runs. - The team_start consent card quoted the policy's fallback rule word "default" instead of the cost clause; approval names its fallback rules and the card treats them as no rule. - team_start's description told the model the person is always asked first, so a manager under an allowing posture reported an approval nobody gave. - After M the keyboard stayed on the teams page; hints said m; the key sheet omitted M m p r d u. - After Raise to $4 the header kept the recurring cap; it reads today's decided raise, the packet the cap gate uses. - A sub-team manager naming its own team in team_post got an unknown-team refusal; it is now told team_post is for the team above. Fixes #1551 Fixes #1552 Fixes #1576 Fixes #1585 Fixes #1591 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…p is not a failure - A task held by the busy-machine gate had already taken its folder lock and checked out its branch, and /stop could not end it until the driver woke. Admission is now checked first: an admitted run keeps the synchronous route, folder, store, copy order (so its refusals still leave no run behind); a held run seeds only its plan store, shows queued with the machine-busy reason, and prepares its folder and copy after admission. A stop on a held run settles it at once through one driver settlement. - After an engine restart an interrupted belt run's plan store stayed live, so its root read working forever with no worker; startup now archives it. - An explicit stop recorded the crew as not kept, so the crew line said failed and offered /redo stronger and the router learned a failure. A stop is now its own outcome with no learning signal and reads stopped. Fixes #1571 Fixes #1554 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…e ledger and the call log - Both image doors chose a file suffix from the provider's declared type and kept a requested .png, so JPEG bytes were saved as .png. One helper in the provider now names an image from its sniffed bytes, falling back to the declared type, and both doors use it. - speak billed through the unnamed auxiliary usage path, so its spend had no role in usage.jsonl, and media HTTP calls wrote no call-log rows. Speech now records under its own role, and every media request writes one start/end call-log pair carrying the provider's cost. Fixes #1586 Fixes #1588 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Every launch of the ssh link teed its stderr to the terminal, so a redial while the TUI held the alternate screen painted ssh's diagnostics over it. Only the first launch writes to the terminal now; a redial's stderr goes to the session's diagnostic tail. Part of #1553 (the stderr item; the keyboard-focus item stays open) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…task titles end on a whole phrase - The / list and /help omitted /land now, /crew cap task <dollars> and /cache clean now, which dispatch and the manual already support. - /compact on a short chat printed the engine's raw sentinel error; it now says nothing to compact yet. - The fallback task title could end on a dangling word ("... each of"); it now stops at the nearest phrase boundary. Part of #1559 (the /new title, key-sheet and /status items stay open) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

A --host redial arrives before the old pipe has detached, so the engine still recorded that dead pipe as the driver and made the returning window a watcher ("another window"). Each surface now carries a stable client id across its redials, and a returning window replaces the recorded driver only when that driver is its own previous connection; another live window keeps the keyboard. Fixes #1553 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

… bad check paths are refused, kept branches reach /land - Every checker in a run drew from one checker-seat tally, so after the first few checks the rest stopped "at its spend ceiling" having spent almost nothing. The ceiling is now kept per check task (the run's task cap is still shared), and a run whose checks did not finish no longer ends done; its landing names the unfinished checks. - A planner-declared per-part check was stored verbatim even when it named a path that does not exist and that the part does not produce (issue_.go). Such a check is now refused at division admission with the check and path named, so the planner rewrites it; nothing guesses a filename. - A run kept on a protected or moved checkout was invisible to /land. Kept, conflicted and aborted run branches are now listed and landable by /land, and editor .orig backups are classified as droppings so they never reach a task branch. Fixes #1572 Fixes #1573 Fixes #1574 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…sed, never used A /task typed on Home in project B's window could open on another window's live engine for project A and commit there, carrying A's uncommitted edits. The keeper, the one place every new conversation is bound, now holds an absolute ask to the workspace the engine answers for and refuses a mismatch visibly ("could not open the conversation in <B> · the engine answered for <A>") before anything is sent. Plain messages and /task are covered as one target path. Part of #1466 (the road that reaches the wrong engine is not yet found; this stops it from running anything there) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…Teams lands on it - The wall built its tiles from the tab strip, which remembers visited and other windows' conversations, so a new window showed another window's conversation as open here. It now draws the front tab and the ones this window holds. - Opening a tile closed the wall but did not leave the page it was opened from, so from Teams it landed back on Teams. Tile opens now go through the shared landing seam. Fixes #1583 Fixes #1577 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…d memory works, long homes dial, check rows show - Model prose placeholders like <id> were parsed as raw HTML and dropped; the shared inline renderer now keeps their source text. - The team naming card's async answer marked the wall dirty but did not wake the paint clock, so it drew only on the next key; it now wakes it. - The composing model list required a music modality; audio-output rows with the music id mark (Lyria) are admitted, plain speech rows are not. - /memory, /remember, /memories and /forget said memory is off on the default hosted engine road because only an in-process agent was consulted. The welcome now says whether the engine remembers, and the hosted surface reaches the engine's memory commands over the wire. - --host under a long codeaf home failed because the ssh control-path check ignored OpenSSH's 17-character temporary suffix; both forms must now fit, or multiplexing is left off. - A plan's check row that repeats its worker's title was deduplicated away from the Tasks column; check rows are never hidden by title. Fixes #1578 Fixes #1581 Fixes #1589 Fixes #1535 Fixes #1580 Part of #1556 (the check-row item; the landing-race item stays open) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

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

The real-binary check showed that archiving an interrupted run's plan store on startup made a task that never ran read done after a restart. A done that nothing earned is worse than the working row it replaced, so the reconcile is removed and #1554 item 1 stays open for a fix that says interrupted. Refs #1554 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

The full session suite showed the refusal is too eager: a declared check that names a package a part will create (go test ./internal/rank) was refused as a missing path, which broke the family-check lift. Guessing which paths a part will produce is not a sound seam, so the refusal is removed and #1573 stays open. Refs #1573 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

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

santoshkumarradha · 2026-09-27T06:36:57Z

Independent read-only review of all 23 commits (origin/dev...4643131c5).

Verdict: 22 ok, 1 minor concern, no blockers.

Minor concern: internal/remote/server.go:1055. Redial ownership is recognised only when the client sends ClientID. Older clients keep the old redial behaviour. This keeps compatibility and adds no regression.

Checked per commit: root cause at the right seam, regressions and security (read-only helper toolset, ssh socket path, key handling), test coverage, and complexity.

Before merge:

Drop Fixes #1571, Fixes #1554 and Fixes #1574 from the squash message. These are partial fixes.
Confirm the 10 internal/tui3 credit-warning test failures also appear on dev without this PR.

🤖 Generated with Claude Code

santoshkumarradha and others added 23 commits September 27, 2026 00:03

changes: the docs-audit batch's change entry

3fb3b04

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

changes: close the batch entry's frontmatter and shorten its title

5c0512c

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

changes: drop the part-check line from the batch entry

4643131

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

Navigation Menu

fix: docs-audit bug batch (headless, reading helpers, held tasks, standing, teams, fan-out checks, media, remote) - #1604

fix: docs-audit bug batch (headless, reading helpers, held tasks, standing, teams, fan-out checks, media, remote)#1604
santoshkumarradha wants to merge 23 commits into
devfrom
fix/docs-audit-batch

santoshkumarradha commented Sep 27, 2026 •

edited

Loading

Uh oh!

santoshkumarradha commented Sep 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

santoshkumarradha commented Sep 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Fixed

Partly fixed (issue stays open)

Needs owner ruling (not changed)

Dropped or left to other PRs

How it was checked

Uh oh!

santoshkumarradha commented Sep 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

santoshkumarradha commented Sep 27, 2026 •

edited

Loading