#8 feat(KAN-108): Jira update agent — slot-based progress comments from Meridian activity

[Anonymous]

Originally created by: Akarsh-Hegde
Originally owned by: Akarsh-Hegde

Summary

• Jira updater agent: fetches in-progress tasks via mcp-atlassian, pulls session data per task from the new Meridian MCP get-task-sessions tool, generates 3–5 bullet-point summaries via hermes AIAgent, and posts timed progress comments back to Jira
• Slot-based scheduler daemon: fires at configurable office-hour slots (default: 1pm and 5pm for 9–17 window with 4h interval); SIGTERM-safe via run_in_executor + asyncio.Event
• Meridian MCP enhancements: 4 new tools — get-task-sessions (AI-classified ticket links + session dimensions tags), get-recent-sessions, get-task-breakdown, get-active-task — all with from_time/to_time ISO 8601 range support
• Dedup: jira_update_log table (migration 015) with UNIQUE index on (task_key, period_start, period_end) prevents double-posting on daemon restart

Architecture

mcp-atlassian → fetch in-progress tasks
Meridian MCP  → get-task-sessions (ticket_links JOIN + session_dimensions tags)
hermes        → one-shot AIAgent bullet-point generation (no tools, max_iterations=1)
mcp-atlassian → jira_add_comment (body param, Markdown)
jira_update_log → dedup + audit (state: pending → sent/failed)

Critical fixes applied (from code review)

1. run_update() dispatched via run_in_executor so its asyncio.run() calls (MCP stdio clients) don't conflict with the daemon's event loop — without this every scheduled slot would crash
2. DB connection wrapped in try/finally — previously leaked on MCP subprocess failure
3. _get_update_id() called unconditionally after UPSERT — lastrowid is unreliable on the conflict path across SQLite versions
4. wait_for(stop.wait()) replaces shield(ensure_future(...)) — eliminates orphaned task accumulation each slot

Test plan

• [x] 40 pytest tests passing — parse functions, db helpers, slot logic, comment formatting (no network/subprocess)
• [x] Dry-run verified end-to-end: python -m agents.jira_updater_daemon --task KAN-108 --dry-run
• [x] Live comment posted to KAN-108 and visible in Jira
• [x] jira_update_log dedup confirmed — second --trigger-now on same window skips re-post

CLI reference

# One-shot (all in-progress tasks)
python -m agents.jira_updater_daemon --trigger-now

# Single task
python -m agents.jira_updater_daemon --task KAN-108

# Dry run
python -m agents.jira_updater_daemon --task KAN-108 --dry-run

# Daemon (blocks, fires on schedule)
python -m agents.jira_updater_daemon

Environment variables required in services/.env

JIRA_URL=https://your-org.atlassian.net
JIRA_EMAIL=you@example.com
JIRA_API_TOKEN=...

🤖 Generated with Claude Code

[Anonymous]

• assigned_to: adityaharishch

[Anonymous]

• assigned_to: Akarsh-Hegde

[Anonymous]

Ticket changed by: Akarsh-Hegde

• status: open --> closed

[Anonymous]

Originally posted by: adityaharishch

🎉 This PR is included in version 1.0.0 🎉

The release is available on:

• v1.0.0
• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀