ghtriage
An experimental GitHub issue triage pipeline with typed labels, confidence gates, redaction, and bounded writes.
What Jev decides
Issue type, product area, severity, missing information, actionability, security sensitivity, beginner suitability, duplication, and input hazards.
ghtriage asks Jev for issue type, area, severity and several review signals, then deterministic policy withholds uncertain labels and guards every GitHub write with allowlists and budgets.
- Jev role
- Classify issue content and estimate triage and safety signals while local code controls every write.
- Input
- A redacted issue title and body, author association, reactions, repository-specific labels, and guardrail questions.
- Output
- Choice distributions, a Score, and Noul probabilities.
- Decision frequency
- One guardrail request and one triage request per issue, with optional repeated consistency checks before a write.
- Fallback
- Withhold low-confidence labels, keep writes disabled unless explicitly enabled, and enforce deterministic redaction, allowlists, caps, and budgets.
How it works
Overview
ghtriage is an early-stage Python project for automatic GitHub issue labeling. One Jev request asks eight questions: Choices for issue type and product area, a severity Score, and Noul checks for missing reproduction details, actionability, security sensitivity, good-first-issue suitability, and possible duplication. A separate Jev guardrail battery screens untrusted issue text before the triage questions run.
Regex-based secret redaction happens before model calls. Label confidence thresholds, repository and label allowlists, per-issue caps, a run-wide write budget, and a never-auto set remain deterministic. Writing is disabled by default. The repository includes frozen issue and red-team fixtures, but it was a one-commit experimental project at review time, and this directory has not reproduced its live evaluations or performance claims.
- Language
- Python
- License
- MIT
- Author
- KalyanM45
Sources
Related projects
Jev Review
A code-review pipeline assembled from small typed judgments instead of one long reviewer prompt.
What Jev decides
What is risky, which files and evidence matter, what could fail, how severe it is, and who should review it.
Distill
Route coding-agent work, reasoning effort, and context retention through bounded Jev decisions.
What Jev decides
Which configured model and effort should handle a call, what evidence deserves attention, whether another attempt is useful, and what context should be retained.
jev-model-router
Route Claude Code model tiers and reasoning effort with one parallel Jev classification.
What Jev decides
Which fast, balanced, or deep tier fits the work, how much reasoning it needs, and whether production risk requires escalation.