Strip user mentions from Addie thread context #422
Merged
Conversation
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
When processing thread context for Claude: - Add getSlackUserWithAddieToken() to look up users using Addie's bot token - Collect all unique user IDs mentioned in thread messages - Look up display names in parallel for efficiency - Replace <@u12345678> with @DisplayName in thread context - Fall back to [someone] only if user lookup fails This allows Claude to see who was mentioned and respond appropriately, while preventing it from outputting raw Slack user IDs. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
bokelley
force-pushed
the
bokelley/fix-readme-links
branch
from
789bb7e to
11f61e4
Compare
🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When Addie is in a channel, it now listens to all messages (not just @mentions). For each message, it: 1. Evaluates if the message warrants a response (quick yes/no check) 2. If yes, generates a proposed response 3. Queues the response for admin approval instead of sending directly When an admin approves a queued item: - The message is sent to Slack automatically - Execution result is recorded (success/failure, message timestamp) This enables human-in-the-loop oversight for proactive channel responses. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add quickEvaluate() method to claude-client using Haiku for fast yes/no/react decisions - Add 'engagement' rule type for configuring when Addie should respond in channels - Load engagement rules from database (with sensible defaults if none exist) - Support 'react' response for greetings - adds 👋 emoji instead of full response - Update admin UI to show and manage engagement rules Engagement rules control: - When to respond with a full message (questions about AdCP, help requests) - When to react with emoji (greetings, welcomes, new members) - When to ignore (casual chat, internal discussions) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Rules can now have a context that determines when they're applied: - NULL (default): Always included in main system prompt - 'engagement': Only for "should I respond?" channel evaluation - 'admin': Only when talking to admin users - 'member': Only when talking to organization members - 'anonymous': Only when talking to anonymous/unlinked users Changes: - Add migration 066_rule_contexts.sql to add context column - Remove 'engagement' from RuleType (it's a context, not a type) - Add RuleContext type and getRulesByContext() method - Update buildSystemPrompt() to only use rules with null context - Update admin UI with separate context filter and selector - Add context badges with distinct colors This separates WHAT a rule does (type) from WHEN it applies (context). 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add two seed rules to the migration: 1. Areas of Expertise (default context - main prompt) - Knowledge rule listing topics Addie is expert in - AdCP, agentic advertising, AgenticAdvertising.org, MCP, ad tech 2. Channel Engagement (engagement context) - Behavior rule for should-respond evaluation - Short, focused on YES/REACT/NO decisions - References the areas of expertise indirectly 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Address review comments: - Change "Areas of Expertise" to "What Addie Can Help With" - Ground capabilities in actual tools (search_docs, search_repos, validate_adagents, etc.) - Be honest about MCP/A2A: can web search but no authoritative docs indexed - Update engagement rules to reference actual capabilities This makes the rules more accurate - Addie's expertise comes from its tools, not from claiming knowledge about topics. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Introduces a proper router architecture that classifies incoming messages using Claude Haiku before determining the execution path: - Router generates execution plans: ignore, react (emoji), clarify, or respond - Routing rules are code-managed (not user-editable) for tool alignment - Rules synced to database on startup for admin visibility - Replaces ad-hoc quickEvaluate with structured routing Execution plans include: - Which tools to use for the response - Whether to ask clarifying questions - Conditional logic based on user context (admin, member, etc.) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Remove routing rules database table and sync logic (unnecessary complexity) - Remove rule context field (router handles engagement logic in code) - Rename migration 066 to just seed the capabilities rule - Router is now purely code-based with no database interaction The router still provides: - Haiku-based message classification (ignore/react/clarify/respond) - Expertise areas mapped to tools - Quick pattern matching for obvious cases - Conditional logic based on user context (admin/member) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
…tate 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Remove rule context filter from admin UI (context feature removed) - Remove duplicated AVAILABLE_TOOLS from router (tools come from ROUTING_RULES) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Resolved conflict in server/src/addie/bolt-app.ts: - Kept both getSlackUserWithAddieToken (for user mention resolution) and getChannelInfo (from main branch) - Kept AddieRouter import for the new routing architecture 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Creates a comprehensive test script for the new Haiku-based router: - Tests quickMatch for simple patterns (no API calls) - Tests full router with live Haiku API for complex scenarios - 26 test cases covering ignore, react, and respond actions - Validates tool selection for different topic areas Run with: npx tsx scripts/test-router.ts --live 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Summary
<@U12345678>) from thread context before passing to Claude[someone]to preserve context without exposing raw IDsThis fixes an issue where Claude would output raw Slack user IDs in responses, causing them to render as blank/broken mentions in Slack (showing "Private user info" tooltip).
Test plan
🤖 Generated with Claude Code