workflow-research

Produces bounded, source-pinned research records for GitHub issues before planning.

Updated Dec 18, 2025
One-click install
npx skills add https://github.com/l0lxl0lw/dotfiles --skill workflow-research-l0lxl0lw
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: workflow-research
Source: https://github.com/l0lxl0lw/dotfiles/tree/main/opencode/skills/workflow/workflow-research
Command: npx skills add https://github.com/l0lxl0lw/dotfiles --skill workflow-research-l0lxl0lw

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Researching a GitHub issue before planning often turns into an unbounded repository tour or a mandatory multi-agent pipeline that wastes time. This Skill answers the few technical questions blocking a concrete plan, verifies claims directly in source, and publishes a compact, hash-pinned research record. ## Core Features & Use Cases - Bounded Investigation: Targets 2-4 minutes for small tasks, tracing entry point to logic to persistence instead of scanning whole files or historical documents. - Scoped Delegation: Uses codebase-locator, codebase-pattern-finder, or codebase-analyzer only when genuinely needed, with zero to two specialists and no mandatory Locate-Patterns-Analyze pipeline. - Source-Pinned Records: Publishes a v2 Research record via handoff.py with stable fact IDs, 5-10 decisive source locations, and file hashes bound by the helper. - Use Case: Given a GitHub issue URL, run /research to produce a verified evidence record with test prerequisites and open questions, then hand off to /plan with the exact comment URL. ## Quick Start Run /research on this GitHub issue URL to produce a bounded, source-verified research record ready for the planning stage.

Frequently Asked Questions about workflow-research

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

FAQPage Schema
How do I research a GitHub issue before writing an implementation plan?

Run /research with the issue URL to load compact context via handoff.py, list the few blocking technical questions, and verify claims in source. It publishes a Research record and returns the comment URL for the /plan step.

How long should codebase research take for a small task?

Small tasks target 2-4 minutes of bounded investigation, tracing entry point to logic to persistence. Avoid reading whole large files or scanning every historical document just because a search returned them.

When should I delegate to codebase-locator or codebase-analyzer agents?

Delegate only when locating code is genuinely unclear or a mechanism, database, or auth question remains. Small work normally needs zero or one specialist, at most two, and never a mandatory Locate-Patterns-Analyze pipeline.

What goes into a source-pinned research record?

The record contains facts with stable IDs, concise verified claims, and 5-10 decisive source locations whose file hashes are bound by handoff.py. It also includes test prerequisites and open technical questions, without pasted transcripts or entire files.

What are the limitations of bounded issue research?

It does not implement code, produce a second technical plan, or advance project status to Planning. Stale facts require targeted source checks rather than automatic full reinvestigation, and scope expansion needs a named question and reason.