issue-tracer

Investigate, root-cause, and resolve software issues with evidence-gated fix workflows.

Updated Sep 14, 2026
One-click install
npx skills add https://github.com/pandejesal/drone-nav-sar --skill issue-tracer-pandejesal
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: issue-tracer
Source: https://github.com/pandejesal/drone-nav-sar/tree/main/.swarm/bundled-skills/issue-tracer
Command: npx skills add https://github.com/pandejesal/drone-nav-sar --skill issue-tracer-pandejesal

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve? Bug fixes often ship as partial patches that pass tests but leave the real defect, its sibling occurrences, and unwired code paths in place. This Skill enforces an evidence-first, full-resolution workflow that takes an issue from intake and reproduction through root-cause localization, independent plan critique, implementation review, and PR-ready closure. ## Core Features & Use Cases - Phased investigation protocol: Intake, reproduction, reasoning-guided localization, fix planning, and gated implementation with mandatory artifacts at every phase. - Independent critic gates: Plan critic, implementation review, and final critic run in fresh contexts to refute the work before closure, with SHA-bound review freshness. - Recurrence-class eradication: Sweeps the codebase for the defect class, dispositions every hit, and installs a guardrail (lint rule, type constraint, runtime assertion, or CI check) proven to catch regressions. - Use Case: Given a GitHub issue reporting a crash, the Skill reproduces it, localizes the fault to the exact line, drafts a reviewed fix plan, implements the minimal patch after approval, sweeps for sibling defects, and produces a PR body mapping every acceptance criterion to captured command output. ## Quick Start Ask the agent to investigate and fully resolve issue number 1849 using the issue-tracer workflow, starting with reproduction and a reviewed fix plan before any code changes.

Frequently Asked Questions about issue-tracer

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I root-cause a bug before fixing it?

Reproduce the failure first with exact commands and captured output, then localize hierarchically from file to function to line using stack traces, search, and git history. Rank candidate causes by written bug-specific explanations, and do not patch until the fault is justified at the line level.

How do I prevent a fixed bug from recurring in a codebase?

Characterize the defect class as a pattern, sweep the repository with concrete search predicates, and disposition every hit. Then install a guardrail such as a lint rule, type constraint, runtime assertion, or CI check, and demonstrate it fails on the original defect.

What is an independent implementation review for AI-generated fixes?

A fresh context receives only the diff, root cause, approved plan, and test results, then adversarially tries to refute the patch by finding inputs or callers where it fails. The reviewer records the exact commit SHA examined, and any later edit invalidates the approval.

Does this workflow work with Claude Code, Codex, and other coding agents?

Yes, the protocol is agent-neutral with an adapter table mapping file-edit, plan, web, and subagent tools for OpenCode, Claude Code, Codex, ZCode, and GitHub coding agents. When subagent delegation is unavailable, documented fallback self-review passes apply.

Why do tests passing not prove a bug fix is correct?

Tests passing only shows the patch is plausible; it can overfit the regression test while leaving the real runtime path broken or unwired. Closure requires a written correctness justification against the root cause plus evidence that all entry points are wired and edge cases are covered.