Imported from zxsun2022/UsingAI (
AGENTS.md). Install upstream withnpx skills add zxsun2022/UsingAI. Copyright stays with the author.
UsingAI Repository Agent Rules
This repository contains AI skills, agent configurations, applications, templates, and a file-first personal knowledge base under docs/.
Mandatory bootstrap
Before writing anywhere in this repository:
- Read this file.
- Read the nearest applicable
AGENTS.mdin the target directory and its ancestors. - Follow the most specific applicable rule.
- Inspect the target path before creating or replacing content.
- Never assume repository rules were loaded automatically by the runtime.
Repository areas
| Path | Purpose |
|---|---|
skills/ |
Versioned reusable AI skills |
agents/ |
Agent configurations and prompts |
apps/ |
Applications and prototypes |
templates/ |
Reusable templates |
docs/ |
Canonical file-first knowledge base |
Git policy
Direct commits to main are allowed for
- new independent records under
docs/inbox/,docs/conversations/,docs/sources/, ordocs/reports/; - append-only audit entries;
- fixing obvious formatting errors in a file created by the same Agent during the same task.
Branch and pull request are required for
- modifying this
AGENTS.md; - modifying any file under
skills/; - changing directory structure or schemas;
- bulk edits or migrations;
- deleting or moving files;
- materially revising existing long-lived knowledge under
docs/knowledge/; - promoting content from
docs/experiments/into a stable area.
Safety boundaries
Agents must not:
- force-push or rewrite Git history;
- delete branches, files, or large record sets without explicit approval;
- commit credentials, cookies, tokens, private keys, or secrets;
- fabricate model identities, source URLs, dates, citations, or unavailable conversation turns;
- present AI interpretation as the user's original words;
- silently overwrite historical records.
Provenance
Every AI-created knowledge record should identify, when available:
- executing provider, product, interface, model, and identity confidence;
- Skill ID and version;
- source completeness;
- material files, pages, repositories, and web sources used;
- processing actions and limitations.
Use explicit values such as unknown, partial, or unverified instead of guessing.
Completion report
After a write, report:
- file path and GitHub URL;
- commit SHA or pull request URL;
- whether the change was committed directly or submitted for review;
- any material missing inputs or limitations.