feat(agents): Phase 10A — YouTube transcript subagent#17
Merged
Conversation
- Add youtube_agent_node: fetches transcript via youtube-transcript-api, oEmbed metadata via YouTube API (no key needed), runs hierarchical LLM summarization (chunk → parallel chunk summaries → final summary) - Add YOUTUBE_HANDOFF_TOOL_NAME + transfer_to_youtube_agent @tool in supervisor - Update supervisor system prompt with mandatory YouTube URL routing rules - Extend orchestrator _should_continue to route to youtube_agent node - Add youtube_context: str field to AxonState for cross-turn save flow - Add youtube_summary_model setting (default: google/gemini-2.5-flash) - Install youtube-transcript-api==0.6.3 + defusedxml==0.7.1 Flow: supervisor detects URL → transfer_to_youtube_agent tool call → youtube_agent_node fetches + summarizes → ToolMessage with JSON payload → supervisor presents title/channel/duration/summary/key_points + asks to save
|
Warning Rate limit exceeded
You’ve run out of usage credits. Purchase more in the billing tab. ⌛ How to resolve this issue?After the wait time has elapsed, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans have higher rate limits than the trial, open-source and free plans. In all cases, we re-allow further reviews after a brief timeout. Please see our FAQ for further information. ℹ️ Review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: ASSERTIVE Plan: Pro Run ID: 📒 Files selected for processing (6)
WalkthroughThis PR introduces a YouTube sub-agent to the multi-agent orchestrator, enabling the supervisor to delegate YouTube video URLs for transcript summarization. A new ChangesYouTube Sub-Agent & Routing
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes Possibly related PRs
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 6
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
backend/app/agents/subagents/youtube_agent.py (1)
1-375: 🛠️ Refactor suggestion | 🟠 Major | 🏗️ Heavy liftSplit this sub-agent into smaller modules before it grows further.
At 375 lines, this file already exceeds the backend limit and mixes URL parsing, transcript I/O, chunking, prompting, model initialization, and node orchestration in one place. Please break the fetch/parse/summarize/schema pieces apart now; otherwise the Phase 10B/10C additions will make this much harder to change safely.
As per coding guidelines, "Always modular: one responsibility per file" and "No file longer than 300 lines — split into smaller modules."
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@backend/app/agents/subagents/youtube_agent.py` around lines 1 - 375, The file is doing multiple responsibilities; split it into smaller modules: 1) create a youtube_fetcher module containing _extract_video_id, _fetch_oembed, _fetch_transcript and the _YOUTUBE_URL_RE constant; 2) create a chunker module with _build_chunks, _CHARS_PER_TOKEN and _CHUNK_SIZE_CHARS; 3) create a summarizer module that holds _get_summary_model, _CHUNK_SUMMARY_SYSTEM, _CHUNK_SUMMARY_USER, _FINAL_SUMMARY_USER, _summarize_chunks, and _build_final_summary (including _MAX_CONCURRENT_SUMMARIES); and 4) keep orchestration in youtube_agent.py with youtube_agent_node, the tool constants, and the small helper _error_response only delegating to the new modules; update imports and adjust calls (e.g., import youtube_fetcher._extract_video_id, chunker._build_chunks, summarizer._summarize_chunks/_build_final_summary) and ensure logging and settings usage remain unchanged so behavior is preserved.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@backend/app/agents/subagents/youtube_agent.py`:
- Around line 218-219: The code currently parses the LLM JSON into a plain dict
and converts key_points with list(), which can turn a string into a char list
and allows arbitrary shape to be persisted to youtube_context; define a Pydantic
v2 model (e.g., YouTubeSummaryPayload with fields summary: str and key_points:
list[str]) near the helpers, then replace the raw json.loads + dict handling in
the function that reads content (the block returning str(data.get("summary",
"")), list(data.get("key_points", []))) to parse/validate via
YouTubeSummaryPayload.model_validate_json(content) (or model_validate after
json.loads) and return the validated summary and key_points; on validation
error, log/raise and avoid persisting corrupted payload. Apply the same change
to the other similar block referenced (lines ~346-355) so both LLM output and
persisted state use the same schema.
- Around line 132-139: The _fetch_transcript function currently offloads
synchronous YouTubeTranscriptApi.get_transcript using asyncio.to_thread which
will exhaust the threadpool under concurrency; replace this by either (A)
switching to an async transcript client that natively supports async calls, or
(B) implement a custom async wrapper that uses httpx.AsyncClient to perform the
same HTTP requests/parameter handling and parse the transcript responses
asynchronously (remove asyncio.to_thread and YouTubeTranscriptApi usage), and if
neither is possible add clear docstring/comments on _fetch_transcript (and where
it’s called) warning about the thread-pool trade-offs and limiting concurrency
(e.g., semaphore) to avoid exhausting the default executor.
In `@backend/app/agents/supervisor.py`:
- Around line 105-113: The supervisor's post-handoff prompt asks "Would you like
me to save this to your library?" even though the supervisor only binds the two
handoff tools and not save_video_transcript; update the supervisor logic (the
code around transfer_to_youtube_agent handling and the handoff tools binding in
supervisor) to either (A) only append the save prompt when save_video_transcript
is actually bound/available in the supervisor's tools set, or (B) bind the
save_video_transcript tool alongside the existing handoff tools so the
affirmative path can be executed; locate references to
transfer_to_youtube_agent, the supervisor tool-binding list, and any code that
formats the final YouTube result (also the similar block around the 164-166
region) and apply the conditional check or tool binding change accordingly.
In `@backend/pyproject.toml`:
- Line 25: The current call to YouTubeTranscriptApi.get_transcript(video_id) is
incompatible with youtube-transcript-api v1.2.0+; replace that call with the new
instance API: construct YouTubeTranscriptApi(), call .fetch(video_id) and then
.to_raw_data() to return the same transcript data (i.e., change
YouTubeTranscriptApi.get_transcript(video_id) to
YouTubeTranscriptApi().fetch(video_id).to_raw_data()). Also update any tests or
callers expecting the old return shape if needed, or alternatively change the
pyproject dependency to youtube-transcript-api<1.2.0 if you prefer to keep the
old static API.
In `@implementationPlan.MD`:
- Around line 261-264: The plan contains conflicting provider/model sources
(hardcoded gpt-4o-mini in youtube_agent vs settings.youtube_summary_model, and
TTS described both as OpenAI direct API and OpenRouter); unify these by choosing
a single configuration surface (e.g., settings.youtube_summary_model for
summarization and settings.tts_provider or settings.tts_api for TTS) and update
all references and descriptions to use those settings; ensure youtube_agent
reads from settings.youtube_summary_model (remove any gpt-4o-mini hardcode) and
the TTS section consistently documents the selected provider (OpenAI vs
OpenRouter) driven by settings.tts_provider so future implementers have one
source of truth.
- Around line 270-302: The markdown fenced-block containing the process flow
(starting with "User pastes YouTube URL" and steps like
"transfer_to_youtube_agent") lacks a language tag and the SQL fences (e.g., code
sections containing "INSERT into video_transcripts" and "INSERT into
audio_entries") are not separated by blank lines; update the flow block to
include a language (e.g., ```text or ```flow) and add a blank line before and
after each SQL fenced block so the SQL fences are isolated, then run
markdownlint to confirm no more fenced-block violations.
---
Outside diff comments:
In `@backend/app/agents/subagents/youtube_agent.py`:
- Around line 1-375: The file is doing multiple responsibilities; split it into
smaller modules: 1) create a youtube_fetcher module containing
_extract_video_id, _fetch_oembed, _fetch_transcript and the _YOUTUBE_URL_RE
constant; 2) create a chunker module with _build_chunks, _CHARS_PER_TOKEN and
_CHUNK_SIZE_CHARS; 3) create a summarizer module that holds _get_summary_model,
_CHUNK_SUMMARY_SYSTEM, _CHUNK_SUMMARY_USER, _FINAL_SUMMARY_USER,
_summarize_chunks, and _build_final_summary (including
_MAX_CONCURRENT_SUMMARIES); and 4) keep orchestration in youtube_agent.py with
youtube_agent_node, the tool constants, and the small helper _error_response
only delegating to the new modules; update imports and adjust calls (e.g.,
import youtube_fetcher._extract_video_id, chunker._build_chunks,
summarizer._summarize_chunks/_build_final_summary) and ensure logging and
settings usage remain unchanged so behavior is preserved.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: ASSERTIVE
Plan: Pro
Run ID: 78a11cb7-164d-43a4-9305-9f6bac327c8e
⛔ Files ignored due to path filters (1)
backend/uv.lockis excluded by!**/*.lock,!**/uv.lock
📒 Files selected for processing (7)
backend/app/agents/orchestrator.pybackend/app/agents/state.pybackend/app/agents/subagents/youtube_agent.pybackend/app/agents/supervisor.pybackend/app/core/config.pybackend/pyproject.tomlimplementationPlan.MD
- Split youtube_agent.py (375 lines) into 3 focused modules: youtube_fetcher.py, youtube_chunker.py, youtube_summarizer.py (each under 170 lines; agent.py now ~140 lines — within 300-line rule) - Add Pydantic validation for LLM JSON output: YouTubeSummaryPayload validates final summary response (prevents string key_points being turned into a char list via list()) YouTubeVideoPayload validates full persisted payload before model_dump_json() - Update transcript API to youtube-transcript-api v1.x instance API: YouTubeTranscriptApi().fetch(video_id).to_raw_data() replaces deprecated static YouTubeTranscriptApi.get_transcript() - Add thread-pool usage docstring to fetch_transcript (asyncio.to_thread note) - Remove premature save CTA from supervisor system prompt: 'Would you like me to save this to your library?' removed since save_video_transcript tool is not yet bound (Phase 10B) - Fix implementationPlan.MD conflicting notes: gpt-4o-mini hardcode → settings.youtube_summary_model (Gemini 2.5 Flash) OpenAI direct TTS → OpenRouter openai/gpt-audio-mini Add language tag to flow fenced block; isolate SQL fences with blank lines
This was referenced May 10, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Flow: supervisor detects URL → transfer_to_youtube_agent tool call → youtube_agent_node fetches + summarizes → ToolMessage with JSON payload → supervisor presents title/channel/duration/summary/key_points + asks to save
Summary by CodeRabbit
New Features
Chores