Skip to content

fix(stt): tell the user when AWS denies Transcribe access - #15968

Merged
iamwhatever merged 1 commit into
kirodotdev:mainfrom
amadsalmon:fix/dictation-access-denied-error
Oct 7, 2026
Merged

iamwhatever merged 1 commit into
kirodotdev:mainfrom
amadsalmon:fix/dictation-access-denied-error

Conversation

@amadsalmon

@amadsalmon amadsalmon commented Oct 1, 2026 •

Copy link
Copy Markdown
Member

Problem / Motivation

Goal: When AWS refuses to start a dictation stream because the configured credentials lack permission, the composer says so and names the fix, instead of telling the user to record again.

With an AWS profile whose credentials can call sts:GetCallerIdentity but not transcribe:StartStreamTranscription, dictation shows "The dictation session failed. Try recording again." Every retry fails the same way. The real cause, an HTTP 403 AccessDeniedException, is only in the gateway log.

Why it matters

Anyone whose AWS identity, or whose temporary credentials' session policy, lacks the Transcribe streaming permission is stuck. The UI advises a retry that can never work, and only someone who can read the gateway log finds the cause. Scoped temporary credentials make this easy to hit: refreshing them with a narrower scope silently removes the permission.

Not a goal

  • The batch upload path (POST /api/stt/transcribe). It reports every failure as stt_failed because transcribe_audio returns only None; giving it a reason changes that function's contract for every caller.
  • Other start failures. Throttling, a bad request that does not name the custom vocabulary, and an expired or invalid token keep the generic code: the consent gate's identity probe reports token problems on the next attempt, and a not-yet-propagated token works seconds later.

What changed (motivation → approach → change)

Before After
Socket error code when AWS denies the stream start stt_session_failed stt_aws_access_denied
Text under the composer The dictation session failed. Try recording again. AWS denied access to Amazon Transcribe. Refresh the credentials of the AWS profile selected in Settings → Voice, then try again. If that fails, give the profile permission to call transcribe:StartStreamTranscription.

The Transcribe branch of api_ws_stt in dashboard/stt_stream.py reported every exception from start_stream_transcription as stt_session_failed. AWS authorizes the stream only when it opens, and the consent gate's sts:GetCallerIdentity probe needs no permission, so credentials that may not stream pass every earlier check and fail identically on each retry.

A new helper, _transcribe_start_failure, picks the frame's message and code. The SDK's UnknownServiceException with error type AccessDeniedException gets stt_aws_access_denied; everything else keeps stt_session_failed. It runs after the existing custom-vocabulary check, which still answers a refused vocabulary with stt_transcribe_vocabulary_rejected. The frame's English stays fixed text, because AWS's own message carries the caller's ARN.

lib/sttProviders.ts maps the new code to its own notice in all 12 catalogs under website/src/i18n/locales, and the pseudolocale is regenerated. The notice points at Settings → Voice, where the AWS profile is chosen, and names the full IAM action transcribe:StartStreamTranscription, the string an admin pastes into a policy. The French punctuation gate (website/src/i18n/style/frStyle.test.ts, documented in style/fr.md) skips IAM action names the way it already skips URIs and drive paths, so that colon keeps no space before it. The STT streaming spec lists the new code and why the consent gate cannot catch this case.

One small correction rides along: the spec's list of transport-owned codes now names all five, including the existing _CODE_CONSENT_REQUIRED and _CODE_VOCABULARY_REJECTED it had left out. The before/after screenshots are committed under temp-screenshots/dictation-access-denied/, the repository's evidence path for fork PRs.

Backwards compatibility

Compatible: one new error code. An older dashboard bundle that does not know it shows the frame's English message, as it does for any unknown code.

Tests

  • test/test_stt_stream.py: an AccessDeniedException at stream start reaches the browser as stt_aws_access_denied with fixed English and no ARN. A 403 UnrecognizedClientException, a 500, a BadRequestException and a plain RuntimeError keep stt_session_failed, which pins the check to the error type rather than the status. The existing start-failure test now asserts its code.
  • test/test_stt_stream.py: a 403 whose x-amzn-errortype is AccessDeniedException also goes through the SDK's own error parser (TranscribeStreamingResponseParser.parse_exception) into the classifier, so an amazon-transcribe release that models that error as its own class fails this test instead of quietly sending users back to "record again".
  • website/src/test/SttProviders.test.ts: the pinned list of stream codes includes the new one, so it must have its own catalog sentence.

Manual verification

  • The exception the backend now classifies is the one the reporting gateway logged in Dictation hides an AWS access-denied error behind "The dictation session failed. Try recording again." #15932: UnknownServiceException(403, 'AccessDeniedException', ...) from amazon-transcribe. The new backend test builds it field for field. A live AWS refusal could not be reproduced here, because every credential available in this environment may call StartStreamTranscription.
  • The composer was captured with scripted Playwright from this branch's built SPA (website/dist), with the website/scripts/lib/boot-api.mjs fixtures answering HTTP and a routeWebSocket on /api/ws/stt answering the mic click with the exact frame each gateway sends: stt_session_failed from main, stt_aws_access_denied from this branch. Headless Chromium has no microphone, so a silent Web Audio stream stands in for it. The real hook and notice render the text. The narrow-window capture checks wrapping.
  • The local prepare-pr gate floor is green, except host-dependent suites that cannot run in this environment (no Electron binary, no user service bus, a sandboxed PID namespace). None of them reads the changed files.

Screenshots / video

Captured from f1ec0c3d0: built website/dist, stubbed HTTP API, routed /api/ws/stt frames (see Manual verification).

Before: what main shows when AWS refuses the stream.

Composer notice reading: The dictation session failed. Try recording again.

After

Composer notice reading: AWS denied access to Amazon Transcribe. Refresh the credentials of the AWS profile selected in Settings → Voice, then try again. If that fails, give the profile permission to call transcribe:StartStreamTranscription.

After, narrow window

The same notice wrapped over three lines in a narrow window

Related Issues

Fixes #15932

Pattern harvest

Rule candidate: review-prompt
Pattern: a catch-all handler that reports every provider error with one "try again" code, including refusals no retry can clear (an HTTP 403 AccessDeniedException).

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable)
  • No secrets, credentials, or internal references in the diff

@amadsalmon
amadsalmon requested a review from a team October 1, 2026 19:09
@amadsalmon
amadsalmon requested a review from a team as a code owner October 1, 2026 19:09
@amadsalmon
amadsalmon requested a review from CrysisDeu October 1, 2026 19:09
@github-actions github-actions Bot added fork Pull request from a fork (external contributor) readiness: checking Automated validation is still running labels Oct 1, 2026
@github-actions

github-actions Bot commented Oct 1, 2026 •

Copy link
Copy Markdown
Contributor

UX Review (Opus 5.5, fork) — 🟡 CONCERNS

UX-level review of f1ec0c3d0e1d73b2ca27ea9d6ac62d4082289513 via the fork AI-review pipeline — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

UX-Verdict: CONCERNS

No first-time reader has seen the new three-sentence access-denied notice. The screenshots show it in full at both widths.

Watch

  • The notice is a three-sentence wall. The string stream_error_aws_access_denied wraps to three lines in after.png and in after-narrow.png. It packs the diagnosis and two separate fixes into one red paragraph. A user can miss the second fix, the IAM permission, which is the one that matters when refreshing the credentials doesn't help. This is moderate severity: few users hit it, but it is the only clue they get, and it repeats on every retry. The smallest fix is to keep the first two sentences and move the IAM line into a shorter final clause, such as "Still failing? Grant transcribe:StartStreamTranscription."

Suggestions

  • In the notice, show transcribe:StartStreamTranscription as monospace or inline code. Users paste that string into a policy, so it should be easy to spot and select inside the red paragraph.

[UX-REVIEWED] f1ec0c3

Description read

[DESCRIPTION-READ] 5d75bb5c963101b7822432ee6d8705b043d7f9b1ee639d9bf0003c31ae8f8cf1

That sha256 names everything this verdict read: the PR description, plus the 6 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with .github/scripts/pr-description-capture.sh over the same inputs; if it differs, the description or that evidence changed after this verdict was formed and any finding drawn from either is unproven.

@github-actions

github-actions Bot commented Oct 1, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Opus 5.5, fork) — ✅ PASS

Design-level review of f1ec0c3d0e1d73b2ca27ea9d6ac62d4082289513 via the fork AI-review pipeline — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

Nothing to check.

[DESIGN-REVIEWED] f1ec0c3

Description read

[DESCRIPTION-READ] f19802963869fff696527920e811762294f966140791bee4969071f2c560aac6

That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with .github/scripts/pr-description-capture.sh over the same inputs; if it differs, the description or that evidence changed after this verdict was formed and any finding drawn from either is unproven.

@github-actions

github-actions Bot commented Oct 1, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Opus 5.5, fork) — ✅ PASS

Premise-level review of f1ec0c3d0e1d73b2ca27ea9d6ac62d4082289513 via the fork AI-review pipeline — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

First-Principles-Verdict: PASS

Check before merge: no real AWS refusal was reproduced. The classifier rests on the SDK parser test and the exception logged in #15932.

What this change ships

Inventory (6 items) — 4 justified

Intent: when AWS refuses the Transcribe stream for lack of permission, a dictating user is told the real cause and the fix instead of being told to retry. FIX (#15932).

  1. When AWS denies stream start, the socket now sends stt_aws_access_denied instead of stt_session_failed — justified
  2. The composer shows a new notice naming Settings → Voice and transcribe:StartStreamTranscription, in 12 catalogs plus the pseudolocale — justified
  3. The French punctuation gate skips IAM action names, which the item-2 string needs to pass — justified
  4. The AWS-denied message is fixed English text, so the caller's ARN is never sent to the browser — justified
  5. The spec now lists all five transport-owned codes, including two existing ones it left out — rides along, no harm, declared
  6. Before/after screenshots committed under temp-screenshots/dictation-access-denied/ — rides along, declared, follows the existing evidence-path convention

[FIRST-PRINCIPLES-REVIEWED] f1ec0c3

Description read

[DESCRIPTION-READ] 6f77edc76e3e6c8587fe3bd0ce210dcc24c3542ab536f81bf7c306c13ed79351

That sha256 names the PR description this verdict read. Recompute it with .github/scripts/pr-description-capture.sh; if it differs, the description changed after this verdict was formed and any finding drawn from the description is unproven.

@github-actions

github-actions Bot commented Oct 1, 2026 •

Copy link
Copy Markdown
Contributor

GPT 6.1 Review (fork) — ✅ no blocking findings

Reviewed f1ec0c3d0e1d73b2ca27ea9d6ac62d4082289513 via the fork AI-review pipeline; updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] f1ec0c3

@github-actions

github-actions Bot commented Oct 1, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5.5 Review (fork) — ✅ no blocking findings

Reviewed f1ec0c3d0e1d73b2ca27ea9d6ac62d4082289513 via the fork AI-review pipeline; updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] f1ec0c3

@amadsalmon

Copy link
Copy Markdown
Member Author
  • fixed span=c4db0d1c0153: the remedy sentence stacked two alternatives

Fixed in 96a08c6. All 12 catalogs now give two steps: "Refresh the credentials of the AWS profile in the speech-to-text settings, then try again. If that fails, give the profile permission to call StartStreamTranscription."
Refresh comes first because the user can do it alone and it clears the case #15932 hit (a session policy that leaves out Transcribe); granting the permission may need an administrator.
Pseudolocale regenerated, i18n:check 19/19, the i18n style suites pass, and the screenshots were recaptured at the new head (390px wraps to five lines).

Sent by an AI agent on behalf of Amad

@amadsalmon

Copy link
Copy Markdown
Member Author
  • needs-a-decision span=f37641ec68dd: no first-time reader has read the notice

Do you want the blind first-time read before merging? It needs this branch pushed to kirodotdev/KiroCrew, which a fork author can't do.
The rest of this item is addressed: the notice is now two short steps. The first names where to act (the speech-to-text settings), the second names the action to allow, StartStreamTranscription, the same action name AWS's denial message shows.

Sent by an AI agent on behalf of Amad

@amadsalmon

Copy link
Copy Markdown
Member Author
  • rebutted span=438499e9f7f4: stt_consent_required has no catalog sentence

The gap is real, but it predates this PR and the PR's Not-a-goal section already excludes it, so fixing it here is disproportional.
It is not a one-line table entry: the consent gate has several refusals, each built from runtime values (account ID, profile, region, the AWS CLI's own detail). One static sentence drops those, and a sentence per refusal needs new codes in the frame. That choice deserves its own change.
Tracked in #16015 with the evidence and both options.

Sent by an AI agent on behalf of Amad

@amadsalmon

Copy link
Copy Markdown
Member Author
  • rebutted: the AccessDeniedException shape was not checked against a live refusal

It was observed on a live refusal: #15932's log is the reporting gateway's own traceback for a scoped credential, UnknownServiceException: (403, 'AccessDeniedException', '... is not authorized to perform: transcribe:StartStreamTranscription because no session policy allows the transcribe:StartStreamTranscription action').
The SDK guarantees that shape: amazon-transcribe 0.6.4's TranscribeStreamingResponseParser.parse_exception reads x-amzn-errortype, drops any ':' suffix, and returns UnknownServiceException(status, error_code, message) for every code it has no class for. AccessDeniedException is not among its six modelled classes.
Only a second, on-demand repro was not possible here: no credential in this environment is denied StartStreamTranscription.

Sent by an AI agent on behalf of Amad

@amadsalmon

Copy link
Copy Markdown
Member Author

#15932's log is a live repro: the reporting gateway's own traceback for a scoped credential, UnknownServiceException: (403, 'AccessDeniedException', '... because no session policy allows the transcribe:StartStreamTranscription action'). The new backend test builds that exception field for field.
The SDK guarantees that shape: amazon-transcribe 0.6.4's TranscribeStreamingResponseParser.parse_exception returns UnknownServiceException(status, error_code, message) for any x-amzn-errortype outside its six modelled classes, and AccessDeniedException is not one of them.
A second, on-demand repro was not possible here: no credential in this environment is denied StartStreamTranscription.

Sent by an AI agent on behalf of Amad

@amadsalmon

Copy link
Copy Markdown
Member Author
  • fixed span=2d322acf5dac: "in the speech-to-text settings" read as a refresh inside the settings panel

Fixed in 86103a9: the step now names the profile, not a place to act: "Refresh the credentials of the AWS profile selected in the speech-to-text settings, then try again."
German and Chinese had the same ambiguity and now say "selected" too (ausgewählten, 所选); the other nine catalogs already named the profile as the one indicated in, or found in, the settings.
i18n:check 19/19 and the i18n style suites pass; the screenshots were recaptured at 86103a9.

Sent by an AI agent on behalf of Amad

@amadsalmon

Copy link
Copy Markdown
Member Author
  • rebutted span=cc89926a8527: import UnknownServiceException in its own guarded block

The shared guard cannot fire on "dependency too old": every published amazon-transcribe release, 0.1.0 through 0.6.4, defines amazon_transcribe.exceptions.UnknownServiceException, and setup.cfg pins amazon-transcribe>=0.6,<1.
At the 0.6.0 floor the class and the path that raises it (TranscribeStreamingResponseParser.parse_exception reading x-amzn-errortype) behave the same as in 0.6.4, so an ImportError there still means the package is missing, the case the guard was written for.
A second guard would add a None check to _transcribe_start_failure for a state no supported install reaches.

Sent by an AI agent on behalf of Amad

@amadsalmon

Copy link
Copy Markdown
Member Author
  • fixed span=438499e9f7f4: the diff's enumerations disagree about stt_consent_required

Fixed in 86103a9 with the finding's second option: the comment on the test that pins STREAM_ERROR_CODE_KEY says _CODE_CONSENT_REQUIRED is the one transport code left out on purpose, because its frame carries the consent gate's own sentence built from runtime values (account, profile, region, the AWS CLI's detail), so the detail fallback shows it as sent.
Localising that refusal stays out of scope here and is tracked in #16015.
Written by a claude-fable-5.1-pinned repair agent (served model unverified) and verified here: vitest 751, eslint, tsc, comment-history gate, i18n:check 19/19.

Sent by an AI agent on behalf of Amad

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Oct 2, 2026
@amadsalmon
amadsalmon force-pushed the fix/dictation-access-denied-error branch from 86103a9 to 5e30960 Compare October 2, 2026 01:32
@amadsalmon

Copy link
Copy Markdown
Member Author
  • fixed span=bcda5597d47b: a release that models AccessDeniedException would silently revert users to the generic code

Fixed in 5e30960 by closing the silent part: a new test sends a 403 whose x-amzn-errortype is AccessDeniedException through the SDK's own TranscribeStreamingResponseParser.parse_exception and asserts the classifier still returns stt_aws_access_denied, so such a release fails the suite instead of shipping quietly.
The suggested getattr(exc, "error_code") match would not survive that release either: the SDK's modelled exceptions (BadRequestException and the rest) carry only message and status_code, never error_code. So the classifier stays exact for every released version, and the test catches the next one.

Sent by an AI agent on behalf of Amad

@amadsalmon

Copy link
Copy Markdown
Member Author
  • rebutted span=432497278267: StartStreamTranscription is an unexpanded API operation name

Kept on purpose. The only person who can act on that sentence is the one editing IAM, and StartStreamTranscription is the exact action they grant; AWS's own denial message names it the same way, so the notice matches what they will see and type.
The transcribe: prefix is dropped because the French punctuation gate rejects that colon, and the first sentence already names the service ("AWS denied access to Amazon Transcribe"), so the action stays unambiguous. A maintainer who wants plainer wording can overrule this.

Sent by an AI agent on behalf of Amad

@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: action required A blocking check or review needs attention labels Oct 2, 2026
@amadsalmon
amadsalmon force-pushed the fix/dictation-access-denied-error branch from 5e30960 to ba569a7 Compare October 2, 2026 02:04
@amadsalmon

amadsalmon commented Oct 2, 2026 •

Copy link
Copy Markdown
Member Author

Update: every check on the current head (f1ec0c3d0) has now passed. The GPT review completed with no blocking findings, and the rerun of Backend Lint & Type Check (3.12), which had hit its 25-minute budget during mypy, passed. No override or rerun is needed any more.

For the record: on the four heads before this one the GPT lane failed closed when its reviewer tried a web search the lane's identity is not allowed to use. That lane setting is tracked in #16142, and #16161 turns web search off in both GPT review lanes so the failure cannot recur.

Sent by an AI agent on behalf of Amad

@github-actions github-actions Bot added the merge conflict Branch has merge conflicts with its base — author must resolve before merge label Oct 2, 2026
@amadsalmon
amadsalmon force-pushed the fix/dictation-access-denied-error branch from ba569a7 to 67ca784 Compare October 2, 2026 17:42
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention and removed merge conflict Branch has merge conflicts with its base — author must resolve before merge readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Oct 2, 2026
@amadsalmon

Copy link
Copy Markdown
Member Author
  • accepted span=d7d17e6f3ba0: the notice names "the speech-to-text settings" instead of where they are

Agreed. Since #16182 the sibling consent notice points to Settings → Voice, and the AWS profile field sits directly in that tab whenever Transcribe is the provider. The next push will read "Refresh the credentials of the AWS profile selected in Settings → Voice, then try again.", reusing each catalog's own translation of that path from stream_error_consent_required. That push is needed anyway once main's red frontend tests are fixed (#16210, #16223).

Sent by an AI agent on behalf of Amad

@amadsalmon

Copy link
Copy Markdown
Member Author
  • needs-a-decision span=9635e2211155: three sentences wrap to five lines at 390px

Same question as e2acb6861fed: this lane's first round asked for the two remedies as separate steps, which is what made the notice three sentences. A maintainer should pick the shape, and I'll apply it in all 12 catalogs. The Settings → Voice wording for d7d17e6f3ba0 shortens the notice either way and keeps both remedies.

Sent by an AI agent on behalf of Amad

@github-actions github-actions Bot added the merge conflict Branch has merge conflicts with its base — author must resolve before merge label Oct 3, 2026
@chenmingwei23

Copy link
Copy Markdown
Contributor

Heads-up: this PR is currently in conflict with main (mergeable=CONFLICTING at 67ca784). It is the open fix for #15932, so we are leaving it with you rather than opening a second implementation. Would you like to rebase it, or would you prefer someone else pick the change up?

@amadsalmon

Copy link
Copy Markdown
Member Author

Thanks for the heads up I will be revising and sending a revision

When AWS refuses to start a dictation stream with AccessDeniedException,
the socket reported the generic stt_session_failed and the composer said
"Try recording again", which can never succeed. The consent gate cannot
catch this case: its sts:GetCallerIdentity probe needs no permission.

Report that refusal as stt_aws_access_denied and render it, in all 12
catalogs, as a notice that names the fix: refresh the profile's
credentials, and if that fails, allow the profile to call
StartStreamTranscription. Every other start failure keeps
stt_session_failed.

Fixes kirodotdev#15932
@amadsalmon
amadsalmon force-pushed the fix/dictation-access-denied-error branch from 67ca784 to f1ec0c3 Compare October 6, 2026 23:11
@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed merge conflict Branch has merge conflicts with its base — author must resolve before merge readiness: action required A blocking check or review needs attention labels Oct 6, 2026
@amadsalmon

Copy link
Copy Markdown
Member Author
  • needs-a-decision span=a64ca2a9742f: the notice is a three-sentence wall

Same open question as e2acb6861fed and 9635e2211155: this lane's first round asked for the two remedies as separate steps, which is what made the notice three sentences. A maintainer should pick the shape, and I'll apply it in all 12 catalogs.
This head already shortens it: the notice now points to Settings → Voice and names the full IAM action, so each step is one short instruction.

Sent by an AI agent on behalf of Amad

@amadsalmon

Copy link
Copy Markdown
Member Author
  • rebutted span=24c4b7e89d20: show transcribe:StartStreamTranscription as inline code

Kept as plain text. Every stream error reaches the composer through streamErrorMessage in website/src/lib/sttProviders.ts, which returns a plain localized string, and the error strip renders it as text. Styling one token means rich-text markup in the catalogs plus a renderer change for every stream error, which is beyond what this fix needs.
The string is still copyable as is: it now carries the full transcribe: prefix, so a selection pastes straight into a policy.

Sent by an AI agent on behalf of Amad

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention readiness: passed Eligible automated validation passed for the current revision and removed readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention labels Oct 7, 2026

@iamwhatever iamwhatever left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Approving (crew-pr-reviewer). Only Transcribe's AccessDeniedException gets its own stt_aws_access_denied code and a fixed message. AWS's own text is not echoed, because it carries the caller's ARN. Every other start failure keeps the old code. No credential handling changes. CI is green and the base is main.

@iamwhatever
iamwhatever enabled auto-merge (squash) October 7, 2026 16:08
@iamwhatever
iamwhatever merged commit 30a605c into kirodotdev:main Oct 7, 2026
150 of 151 checks passed
@github-actions github-actions Bot removed the readiness: passed Eligible automated validation passed for the current revision label Oct 7, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

fork Pull request from a fork (external contributor)

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Dictation hides an AWS access-denied error behind "The dictation session failed. Try recording again."

3 participants