Toolkit
Capability 03 · reviewed 2026-08-01
On this page3 sections
Reusable Skills
A skill packages specialized instructions, reference material, scripts, and assets so an AgentA worker you delegate to: brief it, and it takes steps on its own. A chatbot answers; an agent acts.An LLM that runs tools in a loop toward a goal — it acts, checks the result, decides the next step, and repeats until done.Full definition can load them when a task requires them. Claude Code, Codex, and Copilot now all document first-party skill mechanisms, making the capability more portable than the surrounding product UX suggests.
The Stable Contract
A strong cross-product skill has four parts:
- A discriminating name and description that make activation predictable.
- A lean entrypoint that defines the workflow and its completion criteria.
- Progressively disclosed resources that are loaded only when required.
- Executable helpers for operations that should not be re-generated from prose.
The portable core is the instruction and resource architecture. Invocation controls, allowed-tool syntax, discovery paths, precedence, and compaction behavior remain product-specific.
Skills Are Not Policy
Skills extend what an agent can do and how it approaches a task. Their activation and interpretation still involve model judgment unless the product and configuration force a particular path. Do not use a skill as the sole control for a rule that must never be bypassed.
Use project instructions for stable, always-relevant repository facts. Use a skill for a repeatable task with a beginning, decision points, and a completion condition. Use a hook or external gate for deterministic enforcement.
Portability Strategy
Keep the core procedure vendor-neutral. Put product-specific frontmatter and invocation controls in thin adapters. Test the skill separately in each supported surface because “Agent Skills compatible” does not guarantee identical discovery, argument passing, tool permissions, or persistence through context compaction.
The deeper lenses retain Claude Code examples and April-era implementation detail. Treat them as a concrete reference implementation; validate current product syntax against the source register before copying configuration.
Documented coverage · 2026-08-01
Reusable Skills by product
These classifications are documentation findings, not empirical runtime tests. Surface names are explicit because product-family claims are too coarse.
Claude Code
Anthropic
Claude Code implements Agent Skills with progressive disclosure, manual or model invocation, scripts, resources, and product-specific controls.
Basis: documentedSources: 1
The reviewed skills reference is insufficient to classify hosted discovery and invocation as equivalent to local clients.
Basis: documentedSources: 1
OpenAI Codex
OpenAI
Codex documents reusable skills as the first-party packaging mechanism for repeatable workflows and supporting resources.
Basis: documentedSources: 2
The reviewed build-skills source is insufficient to classify skill discovery and execution across web and cloud as equivalent to local clients.
Basis: documentedSources: 2
GitHub Copilot
GitHub
Copilot documents Agent Skills for specialized instructions, scripts, and resources, with product-specific discovery and precedence.
Source register
Reusable Skills sources
Primary vendor documentation reviewed on the date shown. Links may change after publication.
- Extend Claude with skills (opens in a new tab)
Anthropic · official docs · accessed 2026-08-01
- Build skills (opens in a new tab)
OpenAI · official docs · accessed 2026-08-01
- Overview of customizing GitHub Copilot CLI (opens in a new tab)
GitHub · official docs · accessed 2026-08-01
- GitHub Copilot CLI command reference (opens in a new tab)
GitHub · official docs · accessed 2026-08-01
Comparative implementation depth