life-guide — Reduce Information Gaps

🧭 Help AI see the technical route before it starts building.

life-guide is a general Agent Skill for technical choices, research-heavy tasks, and context-dependent decisions. It helps an AI slow down before it jumps to a tool, library, workflow, or confident recommendation.

When AI is moving too fast

You ask:

“I want a technical animation with particles, captions, and audio reactivity.”

A typical answer may immediately list a few tools. life-guide first checks what can change the route:

🎞️ Is this for real-time interaction or final video rendering?
⚙️ Will it run on the web, in a compositor, inside a 3D tool, or as a desktop app?
🧩 Do you need high frame rate, low latency, maximum visual quality, or repeatable renders?
🔎 Which versions, dependencies, and host limits could block delivery?
🧪 What is the smallest demo that can validate the biggest risk first?

Then it compares Canvas, WebGL, 3D, video-composition, and other relevant routes; checks primary sources; and gives you a practical next step.

What `life-guide` changes

🧭 Expands the option space beyond the first tool or workflow that comes to mind.
🔎 Separates facts, user statements, assumptions, inferences, and unknowns.
📚 Traces decision-changing claims back to official documentation, rules, source code, or original data.
⚖️ Compares cost, time, performance, maintenance, reversibility, and opportunity cost.
🧪 Designs a low-cost validation step before a full build.
🛑 States stop conditions and what new information would change the recommendation.

Ask it:

“Should this be Canvas, WebGL, or a 3D workflow?”

“Can my current stack support it reliably?”

“Is this animation library still worth using?”

“What should I prototype first?”

Simple question? Answer directly. Complex build? Map the route first. High-impact choice? Check the evidence and failure branches before recommending a path.

life-guide does not promise one perfect answer or an exhaustive list. It helps AI and people see more of the real options before committing.

The longer behavior example is in skills/life-guide/references/examples.md.

Installation

Copy the entire life-guide/ directory, not only SKILL.md. Reference files are loaded as needed. If they are missing, some rules cannot be reached. The standard package is in skills/life-guide/. A release ZIP is also available, with life-guide/ as its top-level directory after extraction.

Platform	Location or import method	How to check discovery
Claude Code	Project `.claude/skills/life-guide/`, or user `~/.claude/skills/life-guide/`	Check `/skills`
OpenAI Codex	`.agents/skills/life-guide/`	Invoke `$life-guide` explicitly
Gemini CLI	`.gemini/skills/life-guide/`, or `gemini skills install <path>`	Run `/skills list`; trust the workspace first and reload skills after adding
GitHub Copilot	Project `.github/skills/life-guide/` and supported `.claude/skills` or `.agents/skills`; user `~/.copilot/skills/`	Ask a question and observe loading; `gh skill` may also be available
WorkBuddy local	`life-guide/` inside the host skill directory	Check the skill list for the name and description
WorkBuddy hosted platform	Requires a host-specific package. See `adapters/workbuddy/`	Follow the host's review result
Doubao / Coze	Content mapping, not direct installation of the standard package. See `adapters/`	Not verified

Windows example for a project-level Claude Code installation using Git Bash:

mkdir -p .claude/skills
cp -r /path/to/skills/life-guide .claude/skills/
ls .claude/skills/life-guide/SKILL.md

After installation, try these two boundary checks:

Ask “Which country is Paris in?” The skill should answer directly without asking about career goals or starting an investigation.
Ask “I want to make a technical animation with particles and audio reactivity.” The skill should restate what is known, identify route-changing gaps, and ask prioritized questions instead of immediately naming a tool.

If both behaviors match, discovery and boundaries are working in that host. If they do not, submit the host, version, input, and actual output according to CONTRIBUTING.md.

Verification boundary

Structural checks do not prove outcome quality. Static checks only verify package structure, references, and rule consistency.
The current version has local structural and behavior records. Automatic triggering, cross-host behavior, cross-model behavior, and real-user outcomes remain unverified. Platform status is listed in adapters/, and untested items are labeled as unverified.
Historical behavior records come from 5.x batches. They used one host, one model, manual loading, and human judgment across versions, so they cannot be generalized to the current version or every host.
Tool installation, account login, and external writes require explicit user approval. The skill does not promise that a tool is virus-free or immune from account restrictions.

Evidence and maintenance documents

With/without-skill comparison and boundaries: docs/ab-evidence.md
Release plan and post-release observations: docs/release-plan.md
Behavior cases and run records: tests/
Package boundaries and maintenance rules: docs/architecture/package-boundaries.md

Contributing

Reports about installation failures, missing triggers, unreadable reference files, or unexpected answers are welcome. Include the host, version, input, and actual output. Rule-change discipline is documented in CONTRIBUTING.md.

Language

Chinese documentation is available in README.zh-CN.md.

License

MIT. See LICENSE. The skill package includes the same license file.