Adds fetch_webpage tool that converts web pages to markdown/text format
using Jina AI's reader service. Tool is registered and available for
void agent to access web content.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

Instead of complex thread traversal that was overwriting root values,
now extracts root URI/CID directly from the notification's reply field.
If no reply field exists, treats the post as the root. This ensures
proper thread structure in Bluesky replies.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Added #voidstop check in bsky.py to skip processing mentions containing this keyword
- Checks both the thread context and mention text for #voidstop
- When found, the mention is removed from queue without processing
- Created send_to_void.py CLI tool for quick message testing
- Tool streams void's responses with formatted output for reasoning, tool calls, and responses

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Create dummy tool that allows agent to add custom notes to stream.thought.ack records
- Update acknowledge_post to accept optional note parameter
- Modify bot loop to capture notes from annotate_ack tool calls
- Add note field to stream.thought.ack records (null if not provided)
- Register tool with proper descriptions and tags

This allows void to optionally annotate acknowledgments with contextual observations.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Create acknowledge_post() function that generates stream.thought.ack records
- Use custom NSID instead of standard Bluesky likes for post acknowledgment
- Automatically acknowledge posts after successful replies
- Handle multiple session access patterns in atproto client
- Test confirmed working with actual record creation

This allows void to track engagement without creating visible likes in the UI.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Add create_whitewind_blog_post to register_tools.py TOOL_CONFIGS
- Create whitewind_tool_guide.md explaining tool usage for the agent
- Document when to use blog posts vs regular posts
- Include markdown formatting capabilities and example scenarios

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Create new tool for posting to Whitewind using com.whtwnd.blog.entry lexicon
- Support title, content (markdown), and optional subtitle
- Hardcode theme as github-light and visibility as public
- Generate proper Whitewind URLs (https://whtwnd.com/{handle}/entries/{rkey})
- Follow existing pattern of using environment variables for credentials

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Replace underscore separators with Unicode box-drawing characters
- Add symbols for different log levels (✓ INFO, ⚠ WARNING, ✗ ERROR, etc.)
- Add contextual symbols for different panel types (▶ mentions, ◆ reasoning, ⚙ tools, ✎ posts)
- Indent panel content for better visual hierarchy
- Maintain --simple-logs compatibility for plain text format
- Use vertical bar separators in log timestamps for cleaner alignment

🤖 Generated with [Claude Code](https://claude.ai/code)

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

Implement automatic detachment of user blocks that are attached to the agent
during message processing. This prevents memory accumulation over time.

Features:
- periodic_user_block_cleanup() function to detach all user_ prefixed blocks
- Configurable cleanup interval via --cleanup-interval flag (default: 10 cycles)
- Cleanup can be disabled with --cleanup-interval 0
- Priority handling for cameron.pfiffer.org notifications
- Priority keywords detection (urgent, priority, important, emergency)
- Updated CLAUDE.md with usage examples and Letta docs link

🤖 Generated with [Claude Code](https://claude.ai/code)

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

Replace Rich panel formatting with simple text output using underscores
for titles. This provides a cleaner, more minimal logging style without
external dependencies.

Changes:
- Remove all Rich imports (Panel, Console, Table, Text)
- Replace panel displays with title + underscores format
- Remove --rich command line option
- Update log_with_panel to use simple print statements
- Remove USE_RICH and console global variables

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Display archival memory insert content in Rich panel with preview
- Show success/error messages for archival memory insert results
- Consistent formatting with other tool calls

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Replace plain text reply thread with Rich panels for better formatting
- Hide verbose log messages by changing to debug level
- Filter out usage_statistics and stop_reason message types
- Clean up tool result displays for bluesky posts
- Skip archival_memory_insert result displays (always None)

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Add Rich panel formatting for all logs when --rich flag is used
- Add --reasoning flag to control reasoning display and log level
- Add --simple-logs flag for cleaner log format without timestamps
- Fix archival memory search display with proper Python literal parsing
- Add interleaved notification checking during queue processing
- Filter out and auto-delete like notifications to improve performance
- Remove emojis from log messages for cleaner output
- Add pretty-printed panels for mentions, reasoning, tool calls, and results
- Improve error handling for various notification types

🤖 Generated with [Claude Code](https://claude.ai/code)

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

Updates the prompt to encourage single-post responses by default and only
use threaded replies when truly necessary (complex topics, explicit requests,
or structured answers). This should reduce the tendency to create long
threads for simple responses.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

Adds --no-git flag to bsky.py that skips git operations when exporting
agent state. This prevents the automatic git staging of agent backups
when the bot needs to export state (e.g., during halt operations).

Usage: python bsky.py --no-git

🤖 Generated with [Claude Code](https://claude.ai/code)

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

Add handle deduplication at the start of attach_user_blocks() to prevent
redundant API calls when the same user appears multiple times in a thread.
This helps reduce the spike in block list API calls.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

Increases the maximum number of messages void can post in a single
threaded reply from 4 to 15, allowing for more comprehensive responses
to complex topics.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

Implements a comprehensive queue management tool that allows:
- Viewing queue statistics across all directories (active, errors, no_reply)
- Listing notifications with filtering by handle
- Deleting all notifications from specific handles with dry-run support
- Force deletion option to skip confirmation prompts

Key features:
- Rich console output with formatted tables
- Safety features: dry-run mode and confirmation prompts
- Handles all queue directories (queue/, queue/errors/, queue/no_reply/)
- Proper error handling and validation
- Updated CLAUDE.md with usage examples

This tool helps manage notification queue buildup from bots and spam accounts,
complementing the new ignore_notification tool for proactive filtering.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

Implements a new tool that allows the agent to explicitly ignore notifications
without creating a reply. This is particularly useful for filtering out bot
interactions or spam.

Key features:
- New ignore_notification tool with reason and category tracking
- Conflict detection if agent calls both reply and ignore tools
- Ignored notifications are deleted from queue (not moved to no_reply)
- Proper logging and tracking of ignored notifications

The tool helps the agent make deliberate decisions about which notifications
to engage with, improving interaction quality.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Create queue/no_reply/ directory structure
- Modify process_mention to return "no_reply" when no reply is generated
- Update queue processing to move messages to no_reply/ folder
- Mark no_reply messages as processed to prevent reprocessing
- Update function documentation

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Replace deprecated bluesky_reply instruction with add_post_to_bluesky_reply_thread
- Provide clear guidance on atomic threading approach (multiple calls = thread)
- Add testing mode with --test flag to prevent actual posting and queue deletion
- Implement duplicate detection in queue to prevent reprocessing in test mode
- Update tool detection to terminate bot if deprecated bluesky_reply is used
- Add comprehensive testing mode logging with 🧪 indicators

🤖 Generated with [Claude Code](https://claude.ai/code)

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

Changed process_mention to return False instead of True when no
add_post_to_bluesky_reply_thread tool calls are found. This ensures
that notifications remain in the queue for retry if the agent doesn't
generate a response, rather than being discarded.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Remove bluesky_reply tool to eliminate confusion with threading approach
- Add new add_post_to_bluesky_reply_thread tool for building reply threads atomically
- Each call adds a single post to the reply thread queue
- Handler (bsky.py) manages actual thread state and AT Protocol threading
- Better error recovery: individual post failures don't affect entire thread
- Added TOOL_CHANGELOG.md to document changes for agent migration

🤖 Generated with [Claude Code](https://claude.ai/code)

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

Add quote removal function to strip double quotes from AI responses before posting to prevent quoted text from appearing in posts. Only removes double quotes to preserve contractions and single quotes.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Add user_note_append: Append text to user memory blocks
- Add user_note_replace: Find and replace text in blocks
- Add user_note_set: Completely set block content
- Add user_note_view: View block content

These tools provide fine-grained control over user-specific
memory blocks, allowing the agent to view, append, replace,
or completely rewrite user information as needed.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- user_note_append: Appends text to a user's memory block
- user_note_replace: Replaces specific text within a user's memory block
- user_note_set: Sets the complete content of a user's memory block

All tools handle block creation and attachment automatically.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

When a halt command is received, the bot now properly deletes the queue file
before terminating. This prevents the halt message from being reprocessed
when the bot restarts, which could cause an infinite halt loop.

Changes:
- Added queue_filepath parameter to process_mention()
- Delete queue file and mark notification as processed before exit(0)
- Updated all process_mention() calls to pass the filepath

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Show full reasoning messages without truncation
- Display bluesky_reply messages with content preview and language
- Show archival_memory_search queries clearly
- Display update_block operations with block name and value preview
- Enhanced tool result logging with success/error indicators
- Add specific formatting for known tool types
- Provide meaningful summaries for tool operations

This makes the bot's operation much more transparent and easier to debug.

- Replace loop with first successful candidate selection to avoid sending multiple replies
- Remove iteration logic that was causing duplicate responses when agent had multiple successful tool calls
- Update logging to show when additional candidates are skipped
- Simplify response handling to single attempt instead of retry loop
- Maintain support for both single replies and threaded replies (1-4 messages)

This fixes the issue where agents making multiple successful bluesky_reply tool calls
would send duplicate responses to the same user.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Only process bluesky_reply tool calls with status: "success"
- Skip tool calls with status: "error" to prevent sending failed replies
- Add two-pass processing: collect tool results first, then filter tool calls
- Increase log preview content from 100 to 150 characters for better visibility
- Add clear warning messages when skipping failed or unknown status tool calls
- Prevent sending replies that failed validation (e.g. >300 char limit)

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Replace synchronous agents.messages.create with create_stream
- Add step streaming with max_steps=100 for better responsiveness
- Increase client timeout from 5 to 10 minutes
- Add condensed info-level logging for streaming chunks with emojis
- Add debug-level logging for full chunk details
- Maintain backward compatibility with existing response handling

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Changed order to fetch new notifications from server first
- Then queue new notifications with priority prefixes
- Finally process entire queue (old + new) in priority order

This ensures high-priority notifications (like those from cameron.pfiffer.org)
are discovered and queued immediately rather than waiting behind older messages.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Change default log level from DEBUG to INFO for void_bot logger
- Remove redundant startup messages and tool listings
- Reduce main loop cycle noise (only show stats when messages processed)
- Remove excessive notification processing status messages
- Move session-related logs to DEBUG level in bsky_utils.py
- Clean up user block attachment/detachment logs in tools/blocks.py

Keeps queue processing logs and important status messages while
eliminating noise from normal operation.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Create tools/halt.py with halt_activity function
- Register halt_activity tool in register_tools.py
- Add halt detection in bsky.py process_mention function
- Bot exports agent state and exits when halt tool is called

🤖 Generated with [Claude Code](https://claude.ai/code)

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

Implemented a priority queue system that ensures messages from cameron.pfiffer.org are processed first:

- Modified save_notification_to_queue() to add priority prefix to filenames
- cameron.pfiffer.org messages get "0_" prefix (high priority)
- All other messages get "1_" prefix (normal priority)
- Queue processing sorts by filename, so high priority messages process first
- Added clear logging to indicate when high priority notifications are queued
- Updated function documentation to reflect priority ordering

File naming format:
- High priority: 0_YYYYMMDD_HHMMSS_{reason}_{hash}.json
- Normal priority: 1_YYYYMMDD_HHMMSS_{reason}_{hash}.json

🤖 Generated with [Claude Code](https://claude.ai/code)

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

Fixed NameError: name 'List' is not defined by adding List to the typing imports and removing redundant import inside function.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

Major changes:
- Updated ReplyArgs schema to accept List[str] messages instead of single message
- Added validation for max 4 messages, each max 300 characters
- Updated bluesky_reply function to handle message lists
- Created reply_with_thread_to_notification function in bsky_utils.py
- Added system failure message posting when thread messages fail
- Updated bot processing logic in bsky.py to handle both single and multi-message replies
- Maintained backward compatibility with old message format
- Updated register_tools.py description

Threading behavior:
- Single message: Works exactly like before
- Multiple messages: Creates a linear thread where each reply is a response to the previous one
- Failure handling: Posts "[SYSTEM FAILURE: COULD NOT POST MESSAGE, PLEASE TRY AGAIN]" if a message in the thread fails

🤖 Generated with [Claude Code](https://claude.ai/code)

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

The get_bluesky_feed tool was commented out and not being attached to agents. Uncommented it so it can be registered with agents.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Collect all bluesky_reply tool calls instead of stopping at first one
- Try each reply candidate until one succeeds or all fail
- Remove fallback to text messages - only use tool calls
- Add detailed logging for debugging which candidate worked/failed
- Improves reliability when Letta API timeouts occur

🤖 Generated with [Claude Code](https://claude.ai/code)

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

• Created queue_manager.py for browsing and managing notification queue
• Interactive paging through notifications with rich terminal UI
• Toggle deletion flags, view full notifications, batch delete operations
• Moves deleted files to deleted/ folder and updates processed_notifications.json
• Added language support to post tool with optional lang parameter
• Cleaned up blocks.py by removing temporary sync code

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Updated bluesky_reply tool to accept optional lang parameter (defaults to en-US)
- Modified bsky.py to extract language from tool call and pass to reply function
- Updated bsky_utils.py reply functions to accept and use langs parameter
- Posts are now tagged with the specified language code for better filtering

This allows Void to post in different languages by specifying codes like
'es', 'ja', 'th', etc. when using the bluesky_reply tool.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Created new reply.py tool module with bluesky_reply function
- Added 300 character validation to match Bluesky's limit
- Registered the tool in register_tools.py
- This simple tool allows the Letta agent to indicate when it wants to send a reply

🤖 Generated with [Claude Code](https://claude.ai/code)

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

Removed defensive memory tools entirely:
- Deleted tools/defensive_memory.py
- Removed safe_memory_insert and safe_core_memory_replace from register_tools.py
- Removed related imports and argument schemas

🤖 Generated with [Claude Code](https://claude.ai/code)

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

Added void-cafe feed (at://did:plc:gfrmhdmjvxn2sjedzboeudef/app.bsky.feed.generator/void-cafe) to the list of available named feeds in the GetBlueskyFeedTool.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Reduced parent_height parameter from 80 to 40 to limit deep thread nesting
- Added debug print/exit to inspect thread prompt structure
- Improved thread preview logging to show meaningful content fields
- This helps identify the token-inefficient YAML representation issue

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Add message counters and statistics tracking
- Create separate logger for prompts to reduce noise
- Add numbered cycle tracking and section markers
- Add progress indicators for queue processing
- Track messages per minute and session totals
- Improve logging of agent responses
- Set httpx logging to CRITICAL to remove noise
- Add detailed logging for initialization and main loop flow

Require user confirmation (y/n) before exporting agent state files and running git operations when starting the handler.

🤖 Generated with [Claude Code](https://claude.ai/code)

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

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Remove Literal type that gets converted to FeedName.value in sandbox
- Keep workaround to strip 'FeedName.' prefix if present
- Use plain string type with validation in function body

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Remove feed_uri parameter to prevent agent from using arbitrary URIs
- Add clear error message listing available feed names when invalid name provided
- Update descriptions to explicitly list available feeds for agent guidance
- Simplify function signature and validation logic

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Replace enum class with inline Literal type annotation
- Simplifies code since enum class won't exist in sandboxed tool
- Provides same validation with cleaner implementation

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Move FEED_PRESETS dictionary inside function body
- Tools must be completely self-contained with no external dependencies
- Update Pydantic schema to use hardcoded list instead of dynamic reference

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Fix agent export to save JSON properly instead of binary
- Add agent_archive/ for timestamped backups (git ignored)
- Update agents/void.af to track current state in git
- Add user block attachment/detachment around agent calls
- Extract handles from nested thread data automatically
- Add named feed presets (discover, ai-for-grownups, atmosphere, home)
- Update feed tool to support feed_name parameter with presets

🤖 Generated with [Claude Code](https://claude.ai/code)

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

- Export void agent state to timestamped .af files on startup
- Create agents/ directory with void_latest.af symlink
- Automatically stage agent exports in git
- Export includes all agent configuration, memory blocks, and tools

🤖 Generated with Claude Code

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