feat(router): LLM-generated contextual acknowledgment messages

[zbigniewsobiecki]

Summary

• Replace hardcoded INITIAL_MESSAGES lookups in webhook ack handlers (Trello, GitHub, JIRA) with a single-shot LLM call that generates a short, context-aware acknowledgment message reflecting the actual request content (card name, PR title, comment text)
• Uses the existing progressModel config (defaults to openrouter:google/gemini-2.5-flash-lite) with a 5-second timeout — zero new configuration required
• Gracefully falls back to static INITIAL_MESSAGES on any failure: no model configured, missing API key, LLM error, timeout, or empty output

Changes

File	Change
src/router/ackMessageGenerator.ts	New — generateAckMessage(), context extractors for each provider, LLM call with fallback chain
src/router/trello.ts	Use generateAckMessage() in tryPostTrelloAck()
src/router/github.ts	Use generateAckMessage() in tryPostGitHubAck()
src/router/jira.ts	Use generateAckMessage() in tryPostJiraAck()
tests/unit/router/ackMessageGenerator.test.ts	New — 30 tests covering context extractors + all fallback scenarios
tests/unit/router/{trello,github,jira}.test.ts	Updated mocks for new ack generator module

Fallback chain

generateAckMessage()
  → LLM generates contextual message     ✓ happy path
  → No progressModel configured          → INITIAL_MESSAGES[agentType]
  → No OPENROUTER_API_KEY credential     → INITIAL_MESSAGES[agentType]
  → LLM call fails (network, auth, etc)  → INITIAL_MESSAGES[agentType]
  → LLM call exceeds 5s timeout          → INITIAL_MESSAGES[agentType]
  → LLM returns empty output             → INITIAL_MESSAGES[agentType]
  → Unknown agentType                    → generic "Working on it..."

Test plan

• 30 unit tests for ackMessageGenerator (context extractors, all fallback paths, env var restoration, timeout)
• Existing router handler tests updated and passing (49 tests)
• Full test suite passing (2600 tests)
• TypeScript compilation clean
• Biome lint clean (0 errors)

🤖 Generated with Claude Code