minhtran3124/agent-harness/xia2

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Find out how something actually works here before building on it. Use when the user asks to add a feature to code they have not explained, names a library or API the repo does not already use, or asks how an existing part of the system works — and NOT when the question is which of several approaches to take, which is brainstorming. Produces a specs/<slug>/research-brief.md.

Install

npx vskill@latest install minhtran3124/agent-harness/xia2

SKILL.md

Xia2 — research before code

<HARD-GATE> Do not write code, scaffold, or edit anything except the research brief until the brief is saved and delivered. An explicit “skip research”/“just implement” waiver skips the workflow, but still requires a one-line warning for every Deep signal; never waive this gate yourself. </HARD-GATE>

Decide depth

Read rules/research-depth.md and references/depth-classifier.md. If intake metadata exists, start with its mapping: high-risk → Deep, normal → Standard, tiny → Quick only when every Quick condition passes. Do not independently reclassify risk. Without intake, use the portable classifier. New evidence may raise depth, never lower it; uncertainty is Standard.

Workflow

  1. Check for an explicit waiver and report any Deep warning.
  2. Read available AGENTS.md, CLAUDE.md, and README.md. Search docs/solutions/INDEX.md,

critical-patterns.md, and at most three relevant solution files; scan relevant recent specs/.

  1. Detect the stack and versions from manifests/configuration, then map local reuse, extension

points, tests, and configuration. Do not infer a stack from directory names or stop after one empty search.

  1. At Standard/Deep, search upstream implementations, and version-matched official documentation

when the change has an external surface (adds/upgrades a dependency, integrates an external system, or relies on a version-specific API). Label each finding Local, Upstream, Docs, or Inference; upstream failure is non-blocking. With no external surface, record that in Source Pack as - none (local-only; no external surface) — never leave it silently empty.

  1. Re-run the depth classifier with evidence. State any upgrade and its source.
  2. Fill references/research-brief-template.md, save it to the supplied spec directory (or

specs/research-brief.md), and deliver it. Recommend reuse → adapt upstream → built-in → build, explaining why rejected alternatives lost.

Quick performs local artifact and reuse search only. Standard adds upstream patterns. Deep adds broad local coverage and explicit risk analysis. Official docs, multiple upstream sources, and changelogs are required by external surface, not by depth alone — see rules/research-depth.md §Coverage. Depth sets how broadly to look; surface sets whether to look outside the repo.

Arguments

$ARGUMENTS is the feature to research; ask for it if absent. A caller may specify quick, standard, or deep, but a requested depth that conflicts with a Deep signal must be surfaced.

References

  • rules/research-depth.md — canonical depth policy
  • references/depth-classifier.md — portable signals and decision details
  • references/research-brief-template.md — required output shape

Security

  • Security scanPassed, 95 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
7
7-day trend score
10

Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/minhtran3124/agent-harness/xia2/badge)](https://verified-skill.com/skills/minhtran3124/agent-harness/xia2)

Something wrong with this skill or its page? Tell us.