Compare commits

...

20 Commits

Author SHA1 Message Date
github-actions[bot]
f33d27f836 chore(release): Update version to v1.4.280 2025-08-10 12:34:52 +00:00
Kayvan Sylvan
1694324261 Merge pull request #1686 from ksylvan/0810-fix-openai-streaming-bug
Prevent duplicate text output in OpenAI streaming responses
2025-08-10 05:32:17 -07:00
Changelog Bot
3a3f5c50a8 chore: incoming 1686 changelog entry 2025-08-10 05:27:29 -07:00
Kayvan Sylvan
b1abfd71c2 fix: prevent duplicate text output in OpenAI streaming responses
## CHANGES

- Skip processing of ResponseOutputTextDone events
- Prevent doubled text in stream output
- Add clarifying comment about API behavior
- Maintain delta chunk streaming functionality
- Fix duplicate content issue in responses
2025-08-10 05:24:30 -07:00
github-actions[bot]
f5b7279225 chore(release): Update version to v1.4.279 2025-08-10 12:13:45 +00:00
Kayvan Sylvan
b974e1bfd5 Merge pull request #1685 from ksylvan/0810-fix-gemini-roles-in-sessions
Fix Gemini Role Mapping for API Compatibility
2025-08-10 05:11:16 -07:00
Changelog Bot
8dda68b3b9 chore: incoming 1685 changelog entry 2025-08-10 05:07:06 -07:00
Kayvan Sylvan
33c24e0cb2 fix(gemini): map chat roles to Gemini user/model in convertMessages
CHANGES
- Map assistant role to model per Gemini constraints
- Map system, developer, function, tool roles to user
- Default unrecognized roles to user to preserve instruction context
- Add unit test validating convertMessages role mapping logic
- Import chat package in tests for role constants
2025-08-10 04:55:33 -07:00
github-actions[bot]
8fb0c5b8a8 chore(release): Update version to v1.4.278 2025-08-09 17:26:37 +00:00
Kayvan Sylvan
d82122b624 Merge pull request #1681 from ksylvan/0803-youtube-transcript-lang-fix
Enhance YouTube Support with Custom yt-dlp Arguments
2025-08-09 10:24:09 -07:00
Kayvan Sylvan
f5966af95a docs: update release notes 2025-08-09 10:09:32 -07:00
Changelog Bot
9470ee1655 chore: incoming 1681 changelog entry 2025-08-09 10:06:50 -07:00
Kayvan Sylvan
9a118cf637 refactor: replace custom arg parser with shellquote; precompile regexes
CHANGES
- Precompile regexes for video, playlist, VTT tags, durations.
- Parse yt-dlp additional arguments using shellquote.Split for safety.
- Validate user-provided yt-dlp args and surface quoting errors.
- Reuse compiled regex in GetVideoOrPlaylistId extractions for stability.
- Simplify removeVTTTags by leveraging precompiled VTT tag matcher.
- Parse ISO-8601 durations with precompiled pattern for efficiency.
- Replace inline VTT language regex with cached compiled matcher.
- Remove unused findVTTFiles helper and redundant language checks.
- Add go-shellquote dependency in go.mod and go.sum.
- Reduce allocations by eliminating per-call regexp.MustCompile invocations.
2025-08-09 10:01:24 -07:00
Kayvan Sylvan
d69757908f docs: update release notes 2025-08-08 23:49:55 -07:00
Changelog Bot
30525ef1c0 chore: incoming 1681 changelog entry 2025-08-08 23:44:42 -07:00
Kayvan Sylvan
8414e72545 feat: add smart subtitle language fallback when requested locale unavailable
CHANGES
- Introduce findVTTFilesWithFallback to handle subtitle language absence
- Prefer requested language VTT, gracefully fallback to available alternatives
- Auto-detect downloaded subtitle language and proceed without interruption
- Update yt-dlp processing to use fallback-aware VTT discovery
- Document language fallback behavior and provide usage example
- Return first available VTT when no specific language requested
- Detect language-coded filenames using regex for robust matching
2025-08-08 23:39:12 -07:00
Kayvan Sylvan
caca366511 docs: update YouTube processing documentation for yt-dlp argument precedence control
## CHANGES

- Add user argument precedence over built-in flags
- Document argument order and override behavior
- Include new precedence section with detailed explanation
- Add override examples for language and format
- Update tips section with precedence guidance
- Modify Go code to append user args last
- Add testing tip for subtitle language discovery
- Include practical override use case examples
2025-08-08 23:14:28 -07:00
Kayvan Sylvan
261eb30951 feat: add --yt-dlp-args flag for custom YouTube downloader options
### CHANGES

- Introduce `--yt-dlp-args` flag for advanced control
- Allow passing browser cookies for authentication
- Improve error handling for YouTube rate limits
- Add comprehensive documentation for YouTube processing
- Refactor YouTube methods to accept additional arguments
- Update shell completions to include new flag
2025-08-08 23:03:02 -07:00
Kayvan Sylvan
bdb36ee296 Merge branch 'main' into 0803-youtube-transcript-lang-fix 2025-08-08 16:42:37 -07:00
Kayvan Sylvan
2b195f204d ci: separate release notes generation into dedicated job
## CHANGES

- Move changelog generation to separate workflow job
- Add fallback logic for YouTube subtitle language detection
- Remove changelog commands from main release job
- Create dedicated update_release_notes job with Go setup
- Implement retry mechanism without language specification
- Improve yt-dlp command argument construction flexibility
- Add proper checkout and Go configuration steps
2025-08-03 21:46:24 -07:00
18 changed files with 521 additions and 43 deletions

View File

@@ -1,11 +1,40 @@
# Changelog
## v1.4.280 (2025-08-10)
### PR [#1686](https://github.com/danielmiessler/Fabric/pull/1686) by [ksylvan](https://github.com/ksylvan): Prevent duplicate text output in OpenAI streaming responses
- Fix: prevent duplicate text output in OpenAI streaming responses
- Skip processing of ResponseOutputTextDone events
- Prevent doubled text in stream output
- Add clarifying comment about API behavior
- Maintain delta chunk streaming functionality
## v1.4.279 (2025-08-10)
### PR [#1685](https://github.com/danielmiessler/Fabric/pull/1685) by [ksylvan](https://github.com/ksylvan): Fix Gemini Role Mapping for API Compatibility
- Fix Gemini role mapping to ensure proper API compatibility by converting chat roles to Gemini's user/model format
- Map assistant role to model role per Gemini API constraints
- Map system, developer, function, and tool roles to user role for proper handling
- Default unrecognized roles to user role to preserve instruction context
- Add comprehensive unit tests to validate convertMessages role mapping logic
## v1.4.278 (2025-08-09)
### PR [#1681](https://github.com/danielmiessler/Fabric/pull/1681) by [ksylvan](https://github.com/ksylvan): Enhance YouTube Support with Custom yt-dlp Arguments
- Add `--yt-dlp-args` flag for custom YouTube downloader options with advanced control capabilities
- Implement smart subtitle language fallback system when requested locale is unavailable
- Add fallback logic for YouTube subtitle language detection with auto-detection of downloaded languages
- Replace custom argument parser with shellquote and precompile regexes for improved performance and safety
## v1.4.277 (2025-08-08)
### PR [#1679](https://github.com/danielmiessler/Fabric/pull/1679) by [ksylvan](https://github.com/ksylvan): Add cross-platform desktop notifications to Fabric CLI
- Add cross-platform desktop notifications with secure custom commands
- Integrate notification sending into chat processing workflow
- Integrate notification sending into chat processing workflow
- Add --notification and --notification-command CLI flags and help
- Provide cross-platform providers: macOS, Linux, Windows with fallbacks
- Escape shell metacharacters to prevent injection vulnerabilities
@@ -45,8 +74,7 @@
### Direct commits
- Chore: remove redundant words
Signed-off-by: queryfast <queryfast@outlook.com>
- Remove redundant words from codebase
- Fix typos in t_ patterns
## v1.4.272 (2025-07-28)

View File

@@ -554,6 +554,7 @@ Application Options:
--notification Send desktop notification when command completes
--notification-command= Custom command to run for notifications (overrides built-in
notifications)
--yt-dlp-args= Additional arguments to pass to yt-dlp (e.g. '--cookies-from-browser brave')
Help Options:
-h, --help Show this help message

View File

@@ -1,3 +1,3 @@
package main
var version = "v1.4.277"
var version = "v1.4.280"

Binary file not shown.

View File

@@ -80,6 +80,7 @@ _fabric() {
'(--transcript-with-timestamps)--transcript-with-timestamps[Grab transcript from YouTube video with timestamps]' \
'(--comments)--comments[Grab comments from YouTube video and send to chat]' \
'(--metadata)--metadata[Output video metadata]' \
'(--yt-dlp-args)--yt-dlp-args[Additional arguments to pass to yt-dlp]:yt-dlp args:' \
'(-g --language)'{-g,--language}'[Specify the Language Code for the chat, e.g. -g=en -g=zh]:language:' \
'(-u --scrape_url)'{-u,--scrape_url}'[Scrape website URL to markdown using Jina AI]:url:' \
'(-q --scrape_question)'{-q,--scrape_question}'[Search question using Jina AI]:question:' \

View File

@@ -13,7 +13,7 @@ _fabric() {
_get_comp_words_by_ref -n : cur prev words cword
# Define all possible options/flags
local opts="--pattern -p --variable -v --context -C --session --attachment -a --setup -S --temperature -t --topp -T --stream -s --presencepenalty -P --raw -r --frequencypenalty -F --listpatterns -l --listmodels -L --listcontexts -x --listsessions -X --updatepatterns -U --copy -c --model -m --modelContextLength --output -o --output-session --latest -n --changeDefaultModel -d --youtube -y --playlist --transcript --transcript-with-timestamps --comments --metadata --language -g --scrape_url -u --scrape_question -q --seed -e --wipecontext -w --wipesession -W --printcontext --printsession --readability --input-has-vars --dry-run --serve --serveOllama --address --api-key --config --search --search-location --image-file --image-size --image-quality --image-compression --image-background --suppress-think --think-start-tag --think-end-tag --disable-responses-api --voice --list-gemini-voices --notification --notification-command --version --listextensions --addextension --rmextension --strategy --liststrategies --listvendors --shell-complete-list --help -h"
local opts="--pattern -p --variable -v --context -C --session --attachment -a --setup -S --temperature -t --topp -T --stream -s --presencepenalty -P --raw -r --frequencypenalty -F --listpatterns -l --listmodels -L --listcontexts -x --listsessions -X --updatepatterns -U --copy -c --model -m --modelContextLength --output -o --output-session --latest -n --changeDefaultModel -d --youtube -y --playlist --transcript --transcript-with-timestamps --comments --metadata --yt-dlp-args --language -g --scrape_url -u --scrape_question -q --seed -e --wipecontext -w --wipesession -W --printcontext --printsession --readability --input-has-vars --dry-run --serve --serveOllama --address --api-key --config --search --search-location --image-file --image-size --image-quality --image-compression --image-background --suppress-think --think-start-tag --think-end-tag --disable-responses-api --voice --list-gemini-voices --notification --notification-command --version --listextensions --addextension --rmextension --strategy --liststrategies --listvendors --shell-complete-list --help -h"
# Helper function for dynamic completions
_fabric_get_list() {
@@ -85,7 +85,7 @@ _fabric() {
return 0
;;
# Options requiring simple arguments (no specific completion logic here)
-v | --variable | -t | --temperature | -T | --topp | -P | --presencepenalty | -F | --frequencypenalty | --modelContextLength | -n | --latest | -y | --youtube | -g | --language | -u | --scrape_url | -q | --scrape_question | -e | --seed | --address | --api-key | --search-location | --image-compression | --think-start-tag | --think-end-tag | --notification-command)
-v | --variable | -t | --temperature | -T | --topp | -P | --presencepenalty | -F | --frequencypenalty | --modelContextLength | -n | --latest | -y | --youtube | --yt-dlp-args | -g | --language | -u | --scrape_url | -q | --scrape_question | -e | --seed | --address | --api-key | --search-location | --image-compression | --think-start-tag | --think-end-tag | --notification-command)
# No specific completion suggestions, user types the value
return 0
;;

View File

@@ -95,6 +95,7 @@ complete -c fabric -l transcript -d "Grab transcript from YouTube video and send
complete -c fabric -l transcript-with-timestamps -d "Grab transcript from YouTube video with timestamps"
complete -c fabric -l comments -d "Grab comments from YouTube video and send to chat"
complete -c fabric -l metadata -d "Output video metadata"
complete -c fabric -l yt-dlp-args -d "Additional arguments to pass to yt-dlp (e.g. '--cookies-from-browser brave')"
complete -c fabric -l readability -d "Convert HTML input into a clean, readable view"
complete -c fabric -l input-has-vars -d "Apply variables to user input"
complete -c fabric -l dry-run -d "Show what would be sent to the model without actually sending it"

298
docs/YouTube-Processing.md Normal file
View File

@@ -0,0 +1,298 @@
# YouTube Processing with Fabric
Fabric provides powerful YouTube video processing capabilities that allow you to extract transcripts, comments, and metadata from YouTube videos and playlists. This guide covers all the available options and common use cases.
## Prerequisites
- **yt-dlp**: Required for transcript extraction. Install on MacOS with:
```bash
brew install yt-dlp
```
Or use the package manager of your choice for your operating system.
See the [yt-dlp wiki page](https://github.com/yt-dlp/yt-dlp/wiki/Installation) for your specific installation instructions.
- **YouTube API Key** (optional): Only needed for comments and metadata extraction. Configure with:
```bash
fabric --setup
```
## Basic Usage
### Extract Transcript
Extract a video transcript and process it with a pattern:
```bash
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --pattern summarize
```
### Extract Transcript with Timestamps
Get transcript with timestamps preserved:
```bash
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --transcript-with-timestamps --pattern extract_wisdom
```
### Extract Comments
Get video comments (requires YouTube API key):
```bash
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --comments --pattern analyze_claims
```
### Extract Metadata
Get video metadata as JSON:
```bash
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --metadata
```
## Advanced Options
### Custom yt-dlp Arguments
Pass additional arguments to yt-dlp for advanced functionality. **User-provided arguments take precedence** over built-in fabric arguments, giving you full control:
```bash
# Use browser cookies for age-restricted or private videos
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --yt-dlp-args "--cookies-from-browser brave"
# Override language selection (takes precedence over -g flag)
fabric -g en -y "https://www.youtube.com/watch?v=VIDEO_ID" --yt-dlp-args "--sub-langs es,fr"
# Use specific format
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --yt-dlp-args "--format best"
# Handle rate limiting (slow down requests)
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --yt-dlp-args "--sleep-requests 1"
# Multiple arguments (use quotes)
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --yt-dlp-args "--cookies-from-browser firefox --write-info-json"
# Combine rate limiting with authentication
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --yt-dlp-args "--cookies-from-browser brave --sleep-requests 1"
# Override subtitle format (takes precedence over built-in --sub-format vtt)
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --yt-dlp-args "--sub-format srt"
```
#### Argument Precedence
Fabric constructs the yt-dlp command in this order:
1. **Built-in base arguments** (`--write-auto-subs`, `--skip-download`, etc.)
2. **Language selection** (from `-g` flag): `--sub-langs LANGUAGE`
3. **User arguments** (from `--yt-dlp-args`): **These override any conflicting built-in arguments**
4. **Video URL**
This means you can override any built-in behavior by specifying it in `--yt-dlp-args`.
### Playlist Processing
Process entire playlists:
```bash
# Process all videos in a playlist
fabric -y "https://www.youtube.com/playlist?list=PLAYLIST_ID" --playlist --pattern summarize
# Save playlist videos to CSV
fabric -y "https://www.youtube.com/playlist?list=PLAYLIST_ID" --playlist -o playlist.csv
```
### Language Support
Specify transcript language:
```bash
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" -g es --pattern translate
```
## Combining Options
You can combine multiple YouTube processing options:
```bash
# Get transcript, comments, and metadata
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" \
--transcript \
--comments \
--metadata \
--pattern comprehensive_analysis
```
## Output Options
### Save to File
```bash
# Save output to file
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --pattern summarize -o summary.md
# Save entire session including input
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --pattern summarize --output-session -o full_session.md
```
### Stream Output
Get real-time streaming output:
```bash
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --pattern summarize --stream
```
## Common Use Cases
### Content Analysis
```bash
# Analyze video content for key insights
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --pattern extract_wisdom
# Check claims made in the video
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --pattern analyze_claims
```
### Educational Content
```bash
# Create study notes from educational videos
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --pattern create_study_notes
# Extract key concepts and definitions
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --pattern extract_concepts
```
### Meeting/Conference Processing
```bash
# Summarize conference talks with timestamps
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" \
--transcript-with-timestamps \
--pattern meeting_summary
# Extract action items from recorded meetings
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --pattern extract_action_items
```
### Content Creation
```bash
# Create social media posts from video content
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --pattern create_social_posts
# Generate blog post from video transcript
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" --pattern write_blog_post
```
## Troubleshooting
### Common Issues
1. **"yt-dlp not found"**: Install yt-dlp using pip or your package manager
2. **Age-restricted videos**: Use `--yt-dlp-args "--cookies-from-browser BROWSER"`
3. **No subtitles available**: Some videos don't have auto-generated subtitles
4. **API rate limits**: YouTube API has daily quotas for comments/metadata
5. **HTTP 429 errors**: YouTube is rate limiting subtitle requests
### Error Messages
- **"YouTube is not configured"**: Run `fabric --setup` to configure YouTube API
- **"yt-dlp failed"**: Check video URL and try with `--yt-dlp-args` for authentication
- **"No transcript content found"**: Video may not have subtitles available
- **"HTTP Error 429: Too Many Requests"**: YouTube rate limit exceeded. This is increasingly common. Solutions:
- **Wait 10-30 minutes and try again** (most effective)
- Use longer sleep: `--yt-dlp-args "--sleep-requests 5"`
- Try with browser cookies: `--yt-dlp-args "--cookies-from-browser brave --sleep-requests 5"`
- **Try a different video** - some videos are less restricted
- **Use a VPN** - different IP address may help
- **Try without language specification** - let yt-dlp choose any available language
- **Try English instead** - `fabric -g en` (English subtitles may be less rate-limited)
### Language Fallback Behavior
When you specify a language (e.g., `-g es` for Spanish) but that language isn't available or fails to download:
1. **Automatic fallback**: Fabric automatically retries without language specification
2. **Smart file detection**: If the fallback downloads a different language (e.g., English), Fabric will automatically detect and use it
3. **No manual intervention needed**: The process is transparent to the user
```bash
# Even if Spanish isn't available, this will work with whatever language yt-dlp finds
fabric -g es -y "https://youtube.com/watch?v=VIDEO_ID" --pattern summarize
```
## Configuration
### YAML Configuration
You can set default yt-dlp arguments in your config file (`~/.config/fabric/config.yaml`):
```yaml
ytDlpArgs: "--cookies-from-browser brave --write-info-json"
```
### Environment Variables
Set up your YouTube API key:
```bash
export FABRIC_YOUTUBE_API_KEY="your_api_key_here"
```
## Tips and Best Practices
1. **Use specific patterns**: Choose patterns that match your use case for better results
2. **Combine with other tools**: Pipe output to other commands or save to files for further processing
3. **Batch processing**: Use playlists to process multiple videos efficiently
4. **Authentication**: Use browser cookies for accessing private or age-restricted content
5. **Language support**: Specify language codes for better transcript accuracy
6. **Rate limiting**: If you encounter 429 errors, use `--sleep-requests 1` to slow down requests
7. **Persistent settings**: Set common yt-dlp args in your config file to avoid repeating them
8. **Argument precedence**: Use `--yt-dlp-args` to override any built-in behavior when needed
9. **Testing**: Use `yt-dlp --list-subs URL` to see available subtitle languages before processing
## Examples
### Quick Video Summary
```bash
fabric -y "https://www.youtube.com/watch?v=dQw4w9WgXcQ" --pattern summarize --stream
```
### Detailed Analysis with Authentication
```bash
fabric -y "https://www.youtube.com/watch?v=VIDEO_ID" \
--yt-dlp-args "--cookies-from-browser chrome" \
--transcript-with-timestamps \
--comments \
--pattern comprehensive_analysis \
-o analysis.md
```
### Playlist Processing
```bash
fabric -y "https://www.youtube.com/playlist?list=PLrAXtmRdnEQy6nuLvVUxpDnx4C0823vBN" \
--playlist \
--pattern extract_wisdom \
-o playlist_wisdom.md
```
### Override Built-in Language Selection
```bash
# Built-in language selection (-g es) is overridden by user args
fabric -g es -y "https://www.youtube.com/watch?v=VIDEO_ID" \
--yt-dlp-args "--sub-langs fr,de,en" \
--pattern translate
```
For more patterns and advanced usage, see the main [Fabric documentation](../README.md).

1
go.mod
View File

@@ -37,6 +37,7 @@ require (
require (
github.com/google/go-cmp v0.7.0 // indirect
github.com/gorilla/websocket v1.5.3 // indirect
github.com/kballard/go-shellquote v0.0.0-20180428030007-95032a82bc51 // indirect
)
require (

2
go.sum
View File

@@ -155,6 +155,8 @@ github.com/joho/godotenv v1.5.1 h1:7eLL/+HRGLY0ldzfGMeQkb7vMd0as4CfYvUVzLqw0N0=
github.com/joho/godotenv v1.5.1/go.mod h1:f4LDr5Voq0i2e/R5DDNOoa2zzDfwtkZa6DnEwAbqwq4=
github.com/json-iterator/go v1.1.12 h1:PV8peI4a0ysnczrg+LtxykD8LfKY9ML6u2jnxaEnrnM=
github.com/json-iterator/go v1.1.12/go.mod h1:e30LSqwooZae/UwlEbR2852Gd8hjQvJoHmT4TnhNGBo=
github.com/kballard/go-shellquote v0.0.0-20180428030007-95032a82bc51 h1:Z9n2FFNUXsshfwJMBgNA0RU6/i7WVaAegv3PtuIHPMs=
github.com/kballard/go-shellquote v0.0.0-20180428030007-95032a82bc51/go.mod h1:CzGEWj7cYgsdH8dAjBGEr58BoE7ScuLd+fwFZ44+/x8=
github.com/kevinburke/ssh_config v1.2.0 h1:x584FjTGwHzMwvHx18PXxbBVzfnxogHaAReU4gf13a4=
github.com/kevinburke/ssh_config v1.2.0/go.mod h1:CT57kijsi8u/K/BOFA39wgDQJ9CxiF4nAY/ojJ6r6mM=
github.com/klauspost/cpuid/v2 v2.0.9/go.mod h1:FInQzS24/EEf25PyTYn52gqo7WaD8xa0213Md/qVLRg=

View File

@@ -113,11 +113,11 @@ func processYoutubeVideo(
}
}
if flags.YouTubeTranscriptWithTimestamps {
if transcript, err = registry.YouTube.GrabTranscriptWithTimestamps(videoId, language); err != nil {
if transcript, err = registry.YouTube.GrabTranscriptWithTimestampsWithArgs(videoId, language, flags.YtDlpArgs); err != nil {
return
}
} else {
if transcript, err = registry.YouTube.GrabTranscript(videoId, language); err != nil {
if transcript, err = registry.YouTube.GrabTranscriptWithArgs(videoId, language, flags.YtDlpArgs); err != nil {
return
}
}

View File

@@ -54,6 +54,7 @@ type Flags struct {
YouTubeTranscriptWithTimestamps bool `long:"transcript-with-timestamps" description:"Grab transcript from YouTube video with timestamps and send to chat"`
YouTubeComments bool `long:"comments" description:"Grab comments from YouTube video and send to chat"`
YouTubeMetadata bool `long:"metadata" description:"Output video metadata"`
YtDlpArgs string `long:"yt-dlp-args" yaml:"ytDlpArgs" description:"Additional arguments to pass to yt-dlp (e.g. '--cookies-from-browser brave')"`
Language string `short:"g" long:"language" description:"Specify the Language Code for the chat, e.g. -g=en -g=zh" default:""`
ScrapeURL string `short:"u" long:"scrape_url" description:"Scrape website URL to markdown using Jina AI"`
ScrapeQuestion string `short:"q" long:"scrape_question" description:"Search question using Jina AI"`

View File

@@ -336,6 +336,19 @@ func (o *Client) convertMessages(msgs []*chat.ChatCompletionMessage) []*genai.Co
for _, msg := range msgs {
content := &genai.Content{Parts: []*genai.Part{}}
switch msg.Role {
case chat.ChatMessageRoleAssistant:
content.Role = "model"
case chat.ChatMessageRoleUser:
content.Role = "user"
case chat.ChatMessageRoleSystem, chat.ChatMessageRoleDeveloper, chat.ChatMessageRoleFunction, chat.ChatMessageRoleTool:
// Gemini's API only accepts "user" and "model" roles.
// Map all other roles to "user" to preserve instruction context.
content.Role = "user"
default:
content.Role = "user"
}
if msg.Content != "" {
content.Parts = append(content.Parts, &genai.Part{Text: msg.Content})
}

View File

@@ -4,6 +4,8 @@ import (
"testing"
"google.golang.org/genai"
"github.com/danielmiessler/fabric/internal/chat"
)
// Test buildModelNameFull method
@@ -51,6 +53,30 @@ func TestExtractTextFromResponse(t *testing.T) {
}
}
// Test convertMessages handles role mapping correctly
func TestConvertMessagesRoles(t *testing.T) {
client := &Client{}
msgs := []*chat.ChatCompletionMessage{
{Role: chat.ChatMessageRoleUser, Content: "user"},
{Role: chat.ChatMessageRoleAssistant, Content: "assistant"},
{Role: chat.ChatMessageRoleSystem, Content: "system"},
}
contents := client.convertMessages(msgs)
expected := []string{"user", "model", "user"}
if len(contents) != len(expected) {
t.Fatalf("expected %d contents, got %d", len(expected), len(contents))
}
for i, c := range contents {
if c.Role != expected[i] {
t.Errorf("content %d expected role %s, got %s", i, expected[i], c.Role)
}
}
}
// Test isTTSModel method
func TestIsTTSModel(t *testing.T) {
client := &Client{}

View File

@@ -115,7 +115,11 @@ func (o *Client) sendStreamResponses(
case string(constant.ResponseOutputTextDelta("").Default()):
channel <- event.AsResponseOutputTextDelta().Delta
case string(constant.ResponseOutputTextDone("").Default()):
channel <- event.AsResponseOutputTextDone().Text
// The Responses API sends the full text again in the
// final "done" event. Since we've already streamed all
// delta chunks above, sending it would duplicate the
// output. Ignore it here to prevent doubled results.
continue
}
}
if stream.Err() == nil {

View File

@@ -25,17 +25,33 @@ import (
"time"
"github.com/danielmiessler/fabric/internal/plugins"
"github.com/kballard/go-shellquote"
"google.golang.org/api/option"
"google.golang.org/api/youtube/v3"
)
var timestampRegex *regexp.Regexp
var languageFileRegex *regexp.Regexp
var videoPatternRegex *regexp.Regexp
var playlistPatternRegex *regexp.Regexp
var vttTagRegex *regexp.Regexp
var durationRegex *regexp.Regexp
const TimeGapForRepeats = 10 // seconds
func init() {
// Match timestamps like "00:00:01.234" or just numbers or sequence numbers
timestampRegex = regexp.MustCompile(`^\d+$|^\d{1,2}:\d{2}(:\d{2})?(\.\d{3})?$`)
// Match language-specific VTT files like .en.vtt, .es.vtt, .en-US.vtt, .pt-BR.vtt
languageFileRegex = regexp.MustCompile(`\.[a-z]{2}(-[A-Z]{2})?\.vtt$`)
// YouTube video ID pattern
videoPatternRegex = regexp.MustCompile(`(?:https?:\/\/)?(?:www\.)?(?:youtube\.com\/(?:live\/|[^\/\n\s]+\/\S+\/|(?:v|e(?:mbed)?)\/|(?:s(?:horts)\/)|\S*?[?&]v=)|youtu\.be\/)([a-zA-Z0-9_-]*)`)
// YouTube playlist ID pattern
playlistPatternRegex = regexp.MustCompile(`[?&]list=([a-zA-Z0-9_-]+)`)
// VTT formatting tags like <c.colorE5E5E5>, </c>, etc.
vttTagRegex = regexp.MustCompile(`<[^>]*>`)
// YouTube duration format PT1H2M3S
durationRegex = regexp.MustCompile(`(?i)PT(?:(\d+)H)?(?:(\d+)M)?(?:(\d+)S)?`)
}
func NewYouTube() (ret *YouTube) {
@@ -76,18 +92,14 @@ func (o *YouTube) initService() (err error) {
}
func (o *YouTube) GetVideoOrPlaylistId(url string) (videoId string, playlistId string, err error) {
// Video ID pattern
videoPattern := `(?:https?:\/\/)?(?:www\.)?(?:youtube\.com\/(?:live\/|[^\/\n\s]+\/\S+\/|(?:v|e(?:mbed)?)\/|(?:s(?:horts)\/)|\S*?[?&]v=)|youtu\.be\/)([a-zA-Z0-9_-]*)`
videoRe := regexp.MustCompile(videoPattern)
videoMatch := videoRe.FindStringSubmatch(url)
// Extract video ID using pre-compiled regex
videoMatch := videoPatternRegex.FindStringSubmatch(url)
if len(videoMatch) > 1 {
videoId = videoMatch[1]
}
// Playlist ID pattern
playlistPattern := `[?&]list=([a-zA-Z0-9_-]+)`
playlistRe := regexp.MustCompile(playlistPattern)
playlistMatch := playlistRe.FindStringSubmatch(url)
// Extract playlist ID using pre-compiled regex
playlistMatch := playlistPatternRegex.FindStringSubmatch(url)
if len(playlistMatch) > 1 {
playlistId = playlistMatch[1]
}
@@ -113,17 +125,27 @@ func (o *YouTube) GrabTranscriptForUrl(url string, language string) (ret string,
func (o *YouTube) GrabTranscript(videoId string, language string) (ret string, err error) {
// Use yt-dlp for reliable transcript extraction
return o.tryMethodYtDlp(videoId, language)
return o.GrabTranscriptWithArgs(videoId, language, "")
}
func (o *YouTube) GrabTranscriptWithArgs(videoId string, language string, additionalArgs string) (ret string, err error) {
// Use yt-dlp for reliable transcript extraction
return o.tryMethodYtDlp(videoId, language, additionalArgs)
}
func (o *YouTube) GrabTranscriptWithTimestamps(videoId string, language string) (ret string, err error) {
// Use yt-dlp for reliable transcript extraction with timestamps
return o.tryMethodYtDlpWithTimestamps(videoId, language)
return o.GrabTranscriptWithTimestampsWithArgs(videoId, language, "")
}
func (o *YouTube) GrabTranscriptWithTimestampsWithArgs(videoId string, language string, additionalArgs string) (ret string, err error) {
// Use yt-dlp for reliable transcript extraction with timestamps
return o.tryMethodYtDlpWithTimestamps(videoId, language, additionalArgs)
}
// tryMethodYtDlpInternal is a helper function to reduce duplication between
// tryMethodYtDlp and tryMethodYtDlpWithTimestamps.
func (o *YouTube) tryMethodYtDlpInternal(videoId string, language string, processVTTFileFunc func(filename string) (string, error)) (ret string, err error) {
func (o *YouTube) tryMethodYtDlpInternal(videoId string, language string, additionalArgs string, processVTTFileFunc func(filename string) (string, error)) (ret string, err error) {
// Check if yt-dlp is available
if _, err = exec.LookPath("yt-dlp"); err != nil {
err = fmt.Errorf("yt-dlp not found in PATH. Please install yt-dlp to use YouTube transcript functionality")
@@ -141,30 +163,93 @@ func (o *YouTube) tryMethodYtDlpInternal(videoId string, language string, proces
// Use yt-dlp to get transcript
videoURL := "https://www.youtube.com/watch?v=" + videoId
outputPath := filepath.Join(tempDir, "%(title)s.%(ext)s")
lang_match := language
if len(language) > 2 {
lang_match = language[:2]
}
cmd := exec.Command("yt-dlp",
baseArgs := []string{
"--write-auto-subs",
"--sub-lang", lang_match,
"--skip-download",
"--sub-format", "vtt",
"--quiet",
"--no-warnings",
"-o", outputPath,
videoURL)
}
args := append([]string{}, baseArgs...)
// Add built-in language selection first
if language != "" {
langMatch := language
if len(langMatch) > 2 {
langMatch = langMatch[:2]
}
args = append(args, "--sub-langs", langMatch)
}
// Add user-provided arguments last so they take precedence
if additionalArgs != "" {
additionalArgsList, err := shellquote.Split(additionalArgs)
if err != nil {
return "", fmt.Errorf("invalid yt-dlp arguments: %v", err)
}
args = append(args, additionalArgsList...)
}
args = append(args, videoURL)
cmd := exec.Command("yt-dlp", args...)
var stderr bytes.Buffer
cmd.Stderr = &stderr
if err = cmd.Run(); err != nil {
err = fmt.Errorf("yt-dlp failed: %v, stderr: %s", err, stderr.String())
return
stderrStr := stderr.String()
// Check for specific YouTube errors
if strings.Contains(stderrStr, "429") || strings.Contains(stderrStr, "Too Many Requests") {
err = fmt.Errorf("YouTube rate limit exceeded. Try again later or use different yt-dlp arguments like '--sleep-requests 1' to slow down requests. Error: %v", err)
return
}
if strings.Contains(stderrStr, "Sign in to confirm you're not a bot") || strings.Contains(stderrStr, "Use --cookies-from-browser") {
err = fmt.Errorf("YouTube requires authentication (bot detection). Use --yt-dlp-args '--cookies-from-browser BROWSER' where BROWSER is chrome, firefox, brave, etc. Error: %v", err)
return
}
if language != "" {
// Fallback: try without specifying language (let yt-dlp choose best available)
stderr.Reset()
fallbackArgs := append([]string{}, baseArgs...)
// Add additional arguments if provided
if additionalArgs != "" {
additionalArgsList, parseErr := shellquote.Split(additionalArgs)
if parseErr != nil {
return "", fmt.Errorf("invalid yt-dlp arguments: %v", parseErr)
}
fallbackArgs = append(fallbackArgs, additionalArgsList...)
}
// Don't specify language, let yt-dlp choose
fallbackArgs = append(fallbackArgs, videoURL)
cmd = exec.Command("yt-dlp", fallbackArgs...)
cmd.Stderr = &stderr
if err = cmd.Run(); err != nil {
stderrStr2 := stderr.String()
if strings.Contains(stderrStr2, "429") || strings.Contains(stderrStr2, "Too Many Requests") {
err = fmt.Errorf("YouTube rate limit exceeded. Try again later or use different yt-dlp arguments like '--sleep-requests 1'. Error: %v", err)
} else {
err = fmt.Errorf("yt-dlp failed with language '%s' and fallback. Original error: %s. Fallback error: %s", language, stderrStr, stderrStr2)
}
return
}
} else {
err = fmt.Errorf("yt-dlp failed: %v, stderr: %s", err, stderrStr)
return
}
}
// Find VTT files using cross-platform approach
vttFiles, err := o.findVTTFiles(tempDir, language)
// Try to find files with the requested language first, but fall back to any VTT file
vttFiles, err := o.findVTTFilesWithFallback(tempDir, language)
if err != nil {
return "", err
}
@@ -172,12 +257,12 @@ func (o *YouTube) tryMethodYtDlpInternal(videoId string, language string, proces
return processVTTFileFunc(vttFiles[0])
}
func (o *YouTube) tryMethodYtDlp(videoId string, language string) (ret string, err error) {
return o.tryMethodYtDlpInternal(videoId, language, o.readAndCleanVTTFile)
func (o *YouTube) tryMethodYtDlp(videoId string, language string, additionalArgs string) (ret string, err error) {
return o.tryMethodYtDlpInternal(videoId, language, additionalArgs, o.readAndCleanVTTFile)
}
func (o *YouTube) tryMethodYtDlpWithTimestamps(videoId string, language string) (ret string, err error) {
return o.tryMethodYtDlpInternal(videoId, language, o.readAndFormatVTTWithTimestamps)
func (o *YouTube) tryMethodYtDlpWithTimestamps(videoId string, language string, additionalArgs string) (ret string, err error) {
return o.tryMethodYtDlpInternal(videoId, language, additionalArgs, o.readAndFormatVTTWithTimestamps)
}
func (o *YouTube) readAndCleanVTTFile(filename string) (ret string, err error) {
@@ -303,8 +388,7 @@ func isTimeStamp(s string) bool {
func removeVTTTags(s string) string {
// Remove VTT tags like <c.colorE5E5E5>, </c>, etc.
tagRegex := regexp.MustCompile(`<[^>]*>`)
return tagRegex.ReplaceAllString(s, "")
return vttTagRegex.ReplaceAllString(s, "")
}
// shouldIncludeRepeat determines if repeated content should be included based on time gap
@@ -428,7 +512,7 @@ func (o *YouTube) GrabDuration(videoId string) (ret int, err error) {
durationStr := videoResponse.Items[0].ContentDetails.Duration
matches := regexp.MustCompile(`(?i)PT(?:(\d+)H)?(?:(\d+)M)?(?:(\d+)S)?`).FindStringSubmatch(durationStr)
matches := durationRegex.FindStringSubmatch(durationStr)
if len(matches) == 0 {
return 0, fmt.Errorf("invalid duration string: %s", durationStr)
}
@@ -588,8 +672,9 @@ func (o *YouTube) normalizeFileName(name string) string {
}
// findVTTFiles searches for VTT files in a directory using cross-platform approach
func (o *YouTube) findVTTFiles(dir, language string) ([]string, error) {
// findVTTFilesWithFallback searches for VTT files, handling fallback scenarios
// where the requested language might not be available
func (o *YouTube) findVTTFilesWithFallback(dir, requestedLanguage string) ([]string, error) {
var vttFiles []string
// Walk through the directory to find VTT files
@@ -612,14 +697,28 @@ func (o *YouTube) findVTTFiles(dir, language string) ([]string, error) {
return nil, fmt.Errorf("no VTT files found in directory")
}
// Prefer files with the specified language
// If no specific language requested, return the first file
if requestedLanguage == "" {
return []string{vttFiles[0]}, nil
}
// First, try to find files with the requested language
for _, file := range vttFiles {
if strings.Contains(file, "."+language+".vtt") {
if strings.Contains(file, "."+requestedLanguage+".vtt") {
return []string{file}, nil
}
}
// Return the first VTT file found if no language-specific file exists
// If requested language not found, check if we have any language-specific files
// This handles the fallback case where yt-dlp downloaded a different language
for _, file := range vttFiles {
// Look for any language pattern (e.g., .en.vtt, .es.vtt, etc.)
if languageFileRegex.MatchString(file) {
return []string{file}, nil
}
}
// If no language-specific files found, return the first VTT file
return []string{vttFiles[0]}, nil
}

View File

@@ -199,6 +199,9 @@ schema = 3
[mod."github.com/json-iterator/go"]
version = "v1.1.12"
hash = "sha256-To8A0h+lbfZ/6zM+2PpRpY3+L6725OPC66lffq6fUoM="
[mod."github.com/kballard/go-shellquote"]
version = "v0.0.0-20180428030007-95032a82bc51"
hash = "sha256-AOEdKETBMUC39ln6jBJ9NYdJWp++jV5lSbjNqG3dV+c="
[mod."github.com/kevinburke/ssh_config"]
version = "v1.2.0"
hash = "sha256-Ta7ZOmyX8gG5tzWbY2oES70EJPfI90U7CIJS9EAce0s="

View File

@@ -1 +1 @@
"1.4.277"
"1.4.280"