AGENT_RESOURCE

Curated skills, workflows, references, and routing guidance for agentic development.

---

What This Is

AGENT_RESOURCE is a D-drive-first library of reusable agent skills, workflow templates, references, and source mirrors. D:\agent-resources is a compatibility junction to this canonical root. Agents should read AGENTS.md, README.md, and SKILL-INDEX.md before changing shared skill assets.

Its job is simple: before an agent improvises a complex workflow, it should check this repository, find the right skill by task intent, read that skill's SKILL.md, and use the existing method.

This repository is designed to be public-safe. It should contain reusable instructions, scripts, templates, and provenance notes; it must not contain local account state, API keys, runtime caches, private logs, or generated toolchain payloads.

Start Here

File or folder	Purpose
AGENTS.md	Agent operating contract for shared skill/resource maintenance.
SKILL-INDEX.md	Main routing map: what each skill does, when to use it, and where it lives.
skills/	Curated installable skill collections.
slash-commands/	Reusable command workflows and checklists.
references/	Documentation, guides, and reusable source notes.
templates/	Project and README templates.
tools/	Discovery, validation, public-safety, and maintenance utilities.

repos/ is intentionally ignored by Git. Keep full upstream clones and bulky mirrors local unless a small curated subset is deliberately promoted into skills/, references/, or templates/.

Implicit Skill Routing

Agents should trigger skills from user intent, not only from exact skill names.

1. Classify the task: debugging, README, research audit, frontend, MCP, security, browser, document, game, writing, etc.
2. Search SKILL-INDEX.md and relevant skill frontmatter.
3. Read the matched SKILL.md.
4. Use bundled scripts/references when the skill provides them.
5. Save useful routing lessons to agentmemory or durable docs.

For deterministic discovery across the complete library, use the dependency-free catalog CLI:

python .\tools\skill_catalog.py catalog
python .\tools\skill_catalog.py search "debug failing tests"
python .\tools\skill_catalog.py search "browser automation" --collection vipin --json
python .\tools\skill_catalog.py validate

The validator checks the core constraints from the Agent Skills specification: required frontmatter failures are blocking, while inherited compatibility debt such as non-kebab-case vendor names, duplicate names, invalid UTF-8, and oversized legacy manifests is reported as warnings. Use --strict for a fully conforming collection. Pull requests and direct pushes apply strict validation only to changed SKILL.md files, so upstream debt cannot grow while existing vendored packs remain usable. New branch and tag pushes compare against the remote default branch; an initial default-branch push with no baseline validates the complete library strictly. Relative-path pointer manifests used by vendored packs are resolved safely inside the repository on every platform.

The architecture, upstream patterns, trust boundaries, and verification evidence are recorded in Skill Library Quality Gate.

Examples:

User asks for	Use
"organize my C/D drives", "clean up workstation", "organize D root folders", "keep devtools and wiki in sync"	vipin/workstation-maintenance, then weiping-wiki for public-safe wiki updates. Live file move plans defer recent files, cap batch sizes by default, generate approval packets, and can preflight every batch without moving files. D-root plans can move eligible root directories into D:\_Organized while preserving old paths as junctions. Protected roots include D:\Research, D:\AGENT_RESOURCE, D:\agent-resources, D:\AGENTIC_SCIENCE, D:\devtools, and public devtools exports. Broad approval can cover all currently passing low-risk batches.
"review this paper", "citation check", "experiment audit"	Project ARIS skills first.
README polish, repo presentation, public onboarding	README/documentation skills.
Broken tests, flaky behavior, unfamiliar error	Systematic debugging skills.
Agent memory, MCP, tool routing, wrapper regressions	Agent architecture / MCP skills.
New or updated skill	Skill creator / skill evaluation skills.

Skill Collections

Collection	Use
vipin/	Local custom skills for workstation maintenance, communication, Lark/Feishu, frontend, paper workflows, and publishing.
anthropics/	Anthropic-oriented skills for API, MCP, web app testing, frontend design, and documents.
obra-superpowers/	Engineering workflows such as debugging, TDD, code review, planning, and parallel work.
context-engineering-kit/	Context engineering, DDD/TDD/review/reflexion/kaizen workflows.
trailofbits/	Security, audit, static analysis, and protocol review skills.
ecc/	Large agent harness skill and agent collection.
standalone/	Independent skill packs and references.

Operating Rules

• Keep skills discoverable: SKILL.md frontmatter descriptions should say exactly when the skill should trigger.
• Preserve source and license provenance for imported skill packs.
• Prefer small curated skill folders over dumping whole upstream repositories.
• Do not commit repos/, caches, browser profiles, generated runtime state, auth files, .env, or secrets.
• Audit untracked additions before committing them.
• Use agentmemory for live memory and cross-agent signals; use this repo for reusable skill assets and routing documentation.
• Track provenance review notes for local skill additions, starting with docs/skill-provenance-audit-2026-06-01.md.
• Treat skills/game-studios/ and skills/godogen/ as local-only until their upstream source and redistribution license are confirmed.
• Route nuwa-skill by intent only through the curated MIT-licensed core files committed in skills/standalone/nuwa-skill/.

Public-Safety Checklist

Before changing repo visibility or publishing a release:

git status --short
git ls-files
powershell .\tools\Test-PublicSafety.ps1
powershell .\tools\Test-HistorySafety.ps1
powershell .\tools\Test-PrePushSafety.ps1
powershell .\tools\Install-PrePushHook.ps1

Treat matches as findings until reviewed. The history scan permits a small allowlist of known documentation placeholders such as ghp_your_new_github_token; real credentials must be removed from current files, cleaned from public history, and rotated before publication.

Smoke Tests

python .\tools\skill_catalog.py validate
python -m unittest discover -s .\tools\tests -p "test_*.py"
powershell .\tools\Test-ImplicitSkillRouting.ps1
powershell .\tools\Test-PublicSafety.ps1
powershell .\tools\Test-HistorySafety.ps1
powershell .\tools\Test-PrePushSafety.ps1

These checks validate portable skill metadata, deterministic intent search, public-safety boundaries, Git history, and passive routing metadata. GitHub Actions runs the portable validation and safety subset on every push and pull request with read-only permissions and SHA-pinned actions.

Related Projects

Project	Role
WEIPING_WIKI / vipin-wiki	Public knowledge base, canonical agent operating contract, and compatibility home for historical vipinknowledge links.
devtools-public	Clean public export of local Windows agent launchers and health checks.
agentmemory	Upstream memory and MCP collaboration substrate.

License

Apache-2.0. Imported skill packs may carry their own upstream licenses; preserve those notices where present.