Software Engineer
Issue Triage — Auto-Label, Auto-Assign, Auto-Prioritize
Key Takeaway
Every new GitLab issue gets automatically classified, labeled, assigned to the right agent, and prioritized — the moment it's created. Zero human triage time.
The Problem
Issue triage is project manager work that nobody wants to do but everybody needs done.
New issue lands. Someone reads it. "Is this a bug or a feature request?" Label it. "Is this frontend or backend?" Assign it. "Is this blocking anything?" Prioritize it. "Wait, didn't someone report this already?" Check for duplicates.
At PyratzLabs, we have issues coming in from multiple sources — internal team, beta testers, automated monitoring alerts. Without triage, the issue board becomes a wall of unlabeled, unassigned noise. Things get lost. Duplicates pile up. Critical bugs sit next to cosmetic requests with no distinction.
I was spending 15 minutes a day just reading issue titles and dragging things into the right columns. Fifteen minutes doesn't sound like much. Over a month, that's 5 hours of reading titles and clicking labels.
The Solution
A GitLab webhook triggers our agent on every new issue. The agent reads the title and description, classifies it, labels it, assigns it to the right sub-agent, scores priority, and checks for duplicates. By the time I see the issue, it's already triaged.
The Process
The webhook fires on issue creation. The agent processes it through four stages:
Stage 1: Classification
pythonShow code
CLASSIFICATION_RULES = {
"bug": ["error", "broken", "fail", "crash", "doesn't work", "500", "exception"],
"feature": ["add", "implement", "new", "support for", "ability to", "would be nice"],
"docs": ["documentation", "readme", "typo", "unclear", "how to"],
"infra": ["deploy", "CI", "pipeline", "docker", "server", "scaling", "memory"]
}
def classify_issue(title: str, description: str) -> str:
text = f"{title} {description}".lower()
scores = {}
for category, keywords in CLASSIFICATION_RULES.items():
scores[category] = sum(1 for kw in keywords if kw in text)
# Agent also uses LLM reasoning for ambiguous cases
return max(scores, key=scores.get) if max(scores.values()) > 0 else "triage"
Stage 2: Assignment
pythonShow code
ASSIGNMENT_MAP = {
"frontend": "jack-ui", # Jack's UI sub-agent
"backend": "thom-backend", # Thom's backend sub-agent
"blockchain": "thom-web3", # Thom's web3 sub-agent
"infra": "thom-devops", # Thom's devops sub-agent
"docs": "thom", # Thom handles docs directly
}
def assign_issue(classification: str, changed_areas: list) -> str:
# Detect area from file paths mentioned, or keywords
for area, agent in ASSIGNMENT_MAP.items():
if area in changed_areas or area in classification:
return agent
return "thom" # Default: lead agent triages further
Stage 3: Priority Scoring
pythonShow code
PRIORITY_SIGNALS = {
"P1_blocking": ["blocking", "production down", "can't login", "data loss", "security"],
"P2_important": ["regression", "broken since", "affects multiple", "workaround needed"],
"P3_normal": ["would be nice", "minor", "cosmetic", "enhancement"]
}
def score_priority(title: str, description: str) -> str:
text = f"{title} {description}".lower()
for priority, signals in PRIORITY_SIGNALS.items():
if any(signal in text for signal in signals):
return priority
return "P3_normal" # Default: nice-to-have
Stage 4: Duplicate Detection
bashShow code
# Search existing issues for similar titles
glab issue list --search "$(echo $TITLE | head -c 50)" --state opened -o json
The agent compares the new issue against open issues using title similarity and keyword overlap. If similarity exceeds 70%, it links them:
View details
🔗 Possible duplicate: This issue looks similar to #47
("Login fails after session timeout")
Similarity: 78% — linked for review.
The full output in Telegram:
View details
📋 New Issue Triaged: #52
Title: "Dashboard charts not loading on Safari"
🏷 Labels: bug, frontend
👤 Assigned: @jack-ui
🔴 Priority: P2 (regression — worked before)
🔗 Related: None found
Auto-applied on GitLab. No action needed.
The GitLab labels and assignment are applied via CLI:
bashShow code
glab issue update 52 \
--label "bug,frontend,P2" \
--assignee jack-ui
The Results
| Metric | Before (Manual) | After (Agent) | Improvement |
|---|---|---|---|
| Triage time per issue | 2-3 minutes | 0 (instant) | 100% automated |
| Daily PM triage time | 15 min/day | 0 min/day | 7.5 hours/month saved |
| Label accuracy | 75% (human inconsistency) | 92% (pattern + LLM) | +17% |
| Assignment accuracy | 80% | 95% | +15% |
| Duplicate detection rate | ~20% (caught manually) | ~80% (auto-linked) | 4x improvement |
| Time to first response | Hours (waiting for triage) | Seconds (auto-assigned) | Near-instant |
Try It Yourself
Wire a GitLab webhook to your agent. The classification is basic keyword matching enhanced with LLM reasoning for edge cases. Assignment maps are project-specific — define yours based on who owns what. Duplicate detection is the highest-value piece: even a simple title-similarity check catches 60% of duplicates.
The best project manager is the one you don't notice. Because the work is already done.
Related case studies
CTO
The Engineering Pipeline — From Idea to Production in 24 Hours
From approved idea to production deployment in one day. See how 31 AI agents coordinate across GitLab, Docker, Vercel, and Render to ship with confidence.
DevOps Engineer
CI/CD Pipeline Monitoring — The Agent That Never Sleeps
How an AI agent monitors CI/CD pipelines 24/7 across 5+ GitLab repos, auto-retries flaky jobs, and delivers failure alerts to Telegram — so developers never context-switch to check builds again.
Software Engineer
Creating Merge Requests From Telegram — One Message, Full CI Pipeline
Learn how PyratzLabs creates GitLab merge requests from Telegram in 15 seconds using AI agents. One message triggers MR creation, CI pipeline, and code review — replacing 10 minutes of manual work.
Want results like these?
Start free with your own AI team. No credit card required.