jira

name	jira
description	Fetch Jira tickets with SQLite caching, AI-driven relevance scoring, and structured summarization. Supports raw JQL query, saved filter by ID, or Polaris view. Outputs relevance-scored markdown summaries with mention detection, entity extraction, and adaptive detail levels. Use when the user asks to search, list, or summarize Jira tickets.
version	4.0.0
author	Michael
tags	["jira","jql","search","tickets","query","filter","view","polaris","backlog","sprint","ntuclink","summarize","cache","relevance"]

Jira Skill

When to Use

• User asks to search, list, or summarize Jira tickets
• User wants ticket summaries for daily briefing
• User needs to review active tickets, sprint backlogs, or saved filters
• User asks about recent Jira activity or ticket status

Do NOT use for: creating/updating tickets, fetching non-NTUC Jira instances.

Usage

Default - current user's active tickets

python /a0/usr/skills/jira/scripts/jira.py

Fetch from saved filter and/or Polaris view

python /a0/usr/skills/jira/scripts/jira.py --filter-id 13811 --view-id 10489904

Search tickets by JQL

python /a0/usr/skills/jira/scripts/jira.py --jql 'project = DPD AND status = "In Progress"'

Any combination of JQL, filter, and view

python /a0/usr/skills/jira/scripts/jira.py --jql '...' --filter-id N --view-id N

Force re-fetch and re-summarize everything

python /a0/usr/skills/jira/scripts/jira.py --force

Fast report from cache (no API calls)

python /a0/usr/skills/jira/scripts/jira.py --cached-only

Arguments

Argument	Required	Default	Description
--jql	No	currentUser() active tickets	JQL query string
--filter-id	No	-	Jira saved filter ID (numeric)
--view-id	No	-	Polaris view ID (numeric)
--limit	No	200	Max tickets to return
--offset	No	0	Pagination offset
--cached-only	No	false	Output cached summaries from DB without API calls
--force	No	false	Skip timestamp checks, force full fetch + re-summarize
--output	No	workdir/jira-output.md	Write results to file
--no-cache	No	false	Bypass JQL cache for view resolution

Output

The output .md file is only created after successful completion of all fetching and summarization. If any error occurs, no file is created. This prevents AI agents from reading incomplete results.

Tickets are ordered chronologically (most recently updated first). Relevance scoring is used for filtering, not ordering.

Progress and diagnostics go to stdout (no separate log file).

Output format per ticket:

---
[1/N] jira/DPD-645: Improve event sync to prevent overage
Relevance: 8/10 | Mention: direct
Source: jira | Key: DPD-645 | Status: IN RELEASE QUEUE (Done) | Priority: High | Assignee: Michael Bui
Work Items: DPD-645, DPD-273 | People: Nikhil Grover | Labels: event-sync, pricing-pipeline
blocks: DPD-273 | parent: DPD-644 | Last Updated: 2026-03-19T13:05:25+08:00
[AI summary with adaptive length based on relevance]
---

Monitoring

The script writes progress to stdout. Key markers to look for:

• Fetching progress: [Fetching X/Y] - currently loading tickets from Jira API
• Summarizing progress: Summarizing [X/Y] - AI is generating a summary for a ticket
• Job completed: COMPLETED (N tickets, fetched=M) at the end
• Job completed with errors: COMPLETED WITH ERRORS (N tickets, M error(s)) - run finished but some tickets failed AI summarization
• Job failed: FAILED: error message - fatal error

Important: The output file is only written after ALL summarization completes. Wait for COMPLETED before reading the output file.

Relevance Scoring

Each ticket summary includes a 1-10 relevance score from the AI, guided by:

• User.md context (professional role, domain expertise, priorities)
• mention_type detection (direct/indirect/none)
• Relevance floors: direct >= 7, indirect >= 5, none >= 1
• Adaptive summary length: direct ~200 words, indirect ~100 words, none ~30 words

Mention Type Detection

Automatically classifies each resource as direct, indirect, or none:

• Direct: assignee, reporter, author of comment, @mentioned, user participation (replied)
• Indirect: linked issues, same epic, watcher
• None: no user signal detected
• Strongest signal wins at resource level (1 direct mention in 20 comments = direct)

Handling Results

• Output file is at --output path (default: workdir/jira-output.md)
• Tickets are ordered chronologically (most recently updated first)
• Each ticket block includes: key, status, priority, assignee, AI summary, relevance score, mention type, work items, people, labels, relationships
• Use --cached-only for fast re-reads without API calls