Skip to content

fix: docs#22

Closed
drewdrewthis wants to merge 3 commits intomainfrom
fix/pdocs
Closed

fix: docs#22
drewdrewthis wants to merge 3 commits intomainfrom
fix/pdocs

Conversation

@drewdrewthis
Copy link
Copy Markdown
Collaborator

@drewdrewthis drewdrewthis commented Jun 20, 2025

PDocs can't handle the type aliases from the client, so we have to export everything manually.

PDocs isn't recognizing or handling aliases
PDocs isn't recognizing or handling aliases
@drewdrewthis drewdrewthis self-assigned this Jun 20, 2025
I don't love this, but it works for now
@rogeriochaves
Copy link
Copy Markdown
Contributor

covered by #25

@0xdeafcafe 0xdeafcafe deleted the fix/pdocs branch July 8, 2025 08:57
drewdrewthis added a commit that referenced this pull request Apr 28, 2026
…dges

Closes proposal §4.3 mandate ("auto-transcribe and pass text" when judge
isn't multimodal) and prove-it report row #22 ("judge always attaches
transcripts of audio messages") which was previously deferred.

Changes:
- New private module scenario/voice/_transcribe.py with transcribe_segments(
  recording, provider=None, only_missing=True) — async, concurrent over
  segments, mutates AudioSegment.transcript in place. Per-segment failures
  isolated and logged; missing STT provider warns and no-ops without
  raising. Module is underscore-prefixed because the public symbol
  `transcribe` (a function from stt.py) was already exported and a
  same-named module would shadow it (caught by failing test_stt cases).
- JudgeAgent.call() detects "conversation has audio" by inspecting
  message content parts; if effective_include_audio() returns False on
  an audio conversation, runs transcribe_segments over the executor's
  _voice_recording then enriches messages — replaces audio-only assistant
  messages with text messages carrying the segment transcript before
  building the transcript string.
- 18 new unit tests across test_transcribe.py (6) and
  test_judge_audio_transcribe.py (12) covering: fill-missing, skip-set,
  skip-empty, missing-provider warn, per-segment-isolation, empty-recording,
  audio-detection on input_audio/audio types, message enrichment with and
  without text part, no-mutation guarantee, text-only model auto-detect.
- specs/voice-agents.feature: 3 @Unit AC-15 scenarios.
- test_feature_file_contract.py: 102→105 scenarios, (69,8,25)→(72,8,25).
- prove-it report header updated; row #22 status note added.

Demo verification deferred — OpenAI quota exhausted on this account.
Unit tests prove the wiring; the visible "agent transcript fills in
manifest.json" can be reverified once the demo runs again.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
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.

3 participants