[KongchangAI]
Concept

Skill

In the context of agents, a capability module that is pre-packaged and can be invoked; users trigger it through commands, and the agent uses it to call tools, read data and produce results.

Timeline (last 90 days)

Oct 29

The skills feature can add specific capabilities such as Word, PPT, and Excel operations to the AI like plugins.

Unverified50%
Oct 25

Skills are reusable fixed operation methods that directly operate local software, covering scenarios such as QQ Music, PDF, Markdown, and Mermaid charts.

Unverified50%
Oct 25

Skill is a form of externalized memory, unrelated to model parameters, recording structured, readable and editable Markdown steps, and is more controllable than pure vector memory.

Unverified50%
Oct 25

The necessary condition for creating a skill is that the instruction needs to be reused; optional conditions include involving specific conventions of an individual or team, and involving niche domain knowledge.

Unverified50%
Oct 25

Philippe explicitly does not recommend creating skills for general knowledge, such as designing relational database schemas, because ordinary large models can already handle it very well.

Unverified50%
Oct 18

Nacho Martinez installed more than 100 skills, each taking up about 3,000 to 5,000 tokens.

Unverified50%
Oct 18

When calling the chat completion API, tools and skills often occupy 80% of the token volume of the entire request (about 35,000 to 100,000 tokens).

Unverified50%
Oct 17

A reasonable onboarding path for adopting the Claude Agent SDK is: first get a single agent plus structured output working, then introduce SessionStore for state persistence, and only then split out sub-agents; skills and plugins should be abstracted only when a second reuse need arises.

Unverified50%
Oct 17

In the Claude Agent SDK, skills are positioned as capability units that encapsulate reuse of an agent's internal logic, and plugins are positioned as the channel for integrating external tools or services into the agent system.

Unverified50%
Oct 17

The updates to the Claude Agent SDK in the second half of 2026 focus on five areas: nested sub-agents (with quantity limits), SessionStore, skills, plugins, and structured output plus permission defaults.

Unverified50%

3 more timeline events

All Facts (13)

Unverified

The skills feature can add specific capabilities such as Word, PPT, and Excel operations to the AI like plugins.

50%
Unverified

Skills are reusable fixed operation methods that directly operate local software, covering scenarios such as QQ Music, PDF, Markdown, and Mermaid charts.

50%
Unverified

Skill is a form of externalized memory, unrelated to model parameters, recording structured, readable and editable Markdown steps, and is more controllable than pure vector memory.

50%
Unverified

The necessary condition for creating a skill is that the instruction needs to be reused; optional conditions include involving specific conventions of an individual or team, and involving niche domain knowledge.

50%
Unverified

Philippe explicitly does not recommend creating skills for general knowledge, such as designing relational database schemas, because ordinary large models can already handle it very well.

50%
Unverified

Nacho Martinez installed more than 100 skills, each taking up about 3,000 to 5,000 tokens.

50%
Unverified

When calling the chat completion API, tools and skills often occupy 80% of the token volume of the entire request (about 35,000 to 100,000 tokens).

50%
Unverified

A reasonable onboarding path for adopting the Claude Agent SDK is: first get a single agent plus structured output working, then introduce SessionStore for state persistence, and only then split out sub-agents; skills and plugins should be abstracted only when a second reuse need arises.

50%
Unverified

In the Claude Agent SDK, skills are positioned as capability units that encapsulate reuse of an agent's internal logic, and plugins are positioned as the channel for integrating external tools or services into the agent system.

50%
Unverified

The updates to the Claude Agent SDK in the second half of 2026 focus on five areas: nested sub-agents (with quantity limits), SessionStore, skills, plugins, and structured output plus permission defaults.

50%
Unverified

The same skill has vastly different effects on different queries and at different levels of policy capability.

50%
Unverified

In the agent domain, an "expert" is essentially a combination of pre-configured system prompts and specific tool permissions, while a "skill" is closer to the concept of an MCP or traditional API plugin.

50%
Unverified

In Coze, skills are atomic operations, workflows are process orchestrations, and agents are the complete product form presented externally.

50%

Source Articles