llms.txt@site/static git:20260926.cb3708fA

Immutable. This exact content never changes and is served at /api/v1/blob/4d93984675fed9ce.

# Agent Skill Implementation

> Empirical research into how agent platforms implement Agent Skill loading, management, and presentation. 46 checks across 10 categories, with benchmark skills for cross-platform testing and automated, transcript-cited platform reports.

## Docs

- [Checks](https://agentskillimplementation.com/checks/index.md): The complete catalog of 46 platform behavior checks across 10 categories, with descriptions and rationale for each.
- [Platform Reports](https://agentskillimplementation.com/platforms/index.md): Side-by-side comparison of skill loading verdicts across tested platforms, one table per check category.
- [Guidance](https://agentskillimplementation.com/guidance/index.md): What the findings mean in practice, organized by who acts on them, with each recommendation citing the check that motivates it.
- [Authoring Guidance](https://agentskillimplementation.com/guidance/authoring/index.md): What skill authors should do differently based on observed platform divergence.
- [Installing and Using Skills](https://agentskillimplementation.com/guidance/using/index.md): Where to install skills, what collides with what, and what headless sessions can and cannot do.
- [Distributing and Packaging Skills](https://agentskillimplementation.com/guidance/distributing/index.md): What a catalog, marketplace, or plugin has to get right for a skill to load on every platform it ships to.
- [Notes for Harness Implementers](https://agentskillimplementation.com/guidance/implementing/index.md): Where tested platforms contradict the specification or each other, and the choices that would narrow the gaps.
- [What a Bundled File Costs on Copilot CLI](https://agentskillimplementation.com/guidance/enumeration-cost/index.md): An audit of five public skills repositories against Copilot CLI's activation file list, with per-file, evals, and packaging-metadata costs.
- [Glossary](https://agentskillimplementation.com/guidance/glossary/index.md): Definitions for the terms used throughout the checks, platform reports, and guidance (canary phrase, push harness, pull harness, and more).
- [Antigravity CLI Report](https://agentskillimplementation.com/platforms/antigravity/index.md): Full automated findings for Antigravity CLI (headless), per check: verdict, vehicle, confidence, and observed fallback behavior.
- [Claude Code Report](https://agentskillimplementation.com/platforms/claude-code/index.md): Full automated findings for Claude Code (headless), per check: verdict, vehicle, confidence, and observed fallback behavior.
- [Codex CLI Report](https://agentskillimplementation.com/platforms/codex/index.md): Full automated findings for Codex CLI (headless), per check: verdict, vehicle, confidence, and observed fallback behavior.
- [GitHub Copilot CLI Report](https://agentskillimplementation.com/platforms/copilot/index.md): Full automated findings for GitHub Copilot CLI (headless), per check: verdict, vehicle, confidence, and observed fallback behavior.
- [Home](https://agentskillimplementation.com/index.md): Overview, current status, and links to the GitHub repository for benchmark skills and platform results.

## Tools

- [skill-validator](https://github.com/agent-ecosystem/skill-validator): CLI tool for validating Agent Skills against the spec.
- [Benchmark Skills](https://github.com/agent-ecosystem/agent-skill-implementation/tree/main/benchmark-skills): 45 benchmark fixtures (spec-compliant skills plus deliberate structural and validation edge cases) with canary phrases for cross-platform behavior testing.
- [Platform Template](https://github.com/agent-ecosystem/agent-skill-implementation/blob/main/platform-findings/template.md): Template for recording per-platform behavior findings.
- [benchmark-runner](https://github.com/agent-ecosystem/agent-skill-implementation/tree/main/benchmark-runner): Automated runner that executes the checks against installed harnesses and generates the platform reports from transcript-cited findings.
- [skillxp](https://github.com/agent-ecosystem/skillxp): Harness-agnostic skill invocation engine the runner builds on: installs skills in a fixture, invokes a harness headlessly, locates and parses the transcript, and traces canary phrases.