What Are Agent Skills?

Agent Skills are folders of instructions, scripts and reference files that teach an AI agent how to do a specific task. Each skill has a SKILL.md file with a short name and description, and the agent loads the full instructions only when a task calls for them.

Agent Skills started at Anthropic and became an open standard in December 2025, and many coding agents and tools now read the same format (December 2025, The New Stack). A team can write a skill once, for example for its review checklist or release process, and use it across agents.

What Changed in Agent Skills in 2026

Trend: code review tools read skills

• GitHub Copilot code review support for Agent Skills and MCP became generally available, with project skills read from directories such as .github/skills (July 2026, GitHub’s changelog).
• Qodo added a Skills page to manage how repository skills apply in review, with counts of triggered reviews, detected violations and merged violations per skill (June 2026, Qodo’s changelog).

Trend: review skills ship through agent marketplaces

• Qodo’s Agentic Toolbox became generally available as a Claude plugin, a Codex plugin and a Kiro Power, bringing its review and context skills into those agents (September 2026, Qodo’s changelog).

How Agent Skills Work

Agent Skills use progressive disclosure, so an agent can have many skills installed without filling its context window (last checked October 2026, the Agent Skills specification):

1. Discovery. At startup, the agent loads only each skill’s name and description, about 100 tokens per skill.
2. Activation. When a task matches a skill’s description, the agent reads the full SKILL.md instructions, which the specification recommends keeping under 5,000 tokens.
3. Execution. The agent follows the instructions and loads scripts, references or assets from the skill folder only when it needs them.

Agent Skills File Structure

A skill is a directory named after the skill, with a SKILL.md file at its root:

api-review/
  SKILL.md
  scripts/
  references/
  assets/

SKILL.md starts with YAML frontmatter, followed by the instructions in Markdown:

---
name: api-review
description: Reviews REST API changes against the team's API guidelines. Use when a change adds or modifies an HTTP endpoint.
---

# API review
1. Check every new endpoint requires authentication.
2. Check error responses use the shared error format.
3. Check new fields are documented in the OpenAPI file.

The specification’s main fields:

• name is required, up to 64 characters, lowercase letters, numbers and hyphens, and must match the directory name.
• description is required, up to 1,024 characters, and says what the skill does and when to use it.
• license, compatibility and metadata are optional.
• allowed-tools is optional and experimental, and lists tools the skill may use.

The specification also recommends keeping SKILL.md under 500 lines and moving detail into reference files.

Agent Skills vs AGENTS.md, MCP and Rules

See AGENTS.md and MCP servers for code review for more on those formats.

Examples of Agent Skills

• Code review checklists, such as API, security or accessibility reviews.
• Release and deployment steps, such as preparing a changelog and tagging a release.
• Framework conventions, such as how the team writes database migrations.
• Document and data tasks, such as filling a template or analyzing a spreadsheet.
• Context retrieval, such as finding the owner of a service or the history of a module.

How to Write a Useful Skill

1. Pick one task the team repeats and explains often.
2. Write a description that says what the skill does and exactly when to use it, since that’s what the agent matches on.
3. Keep the instructions short and ordered, and move long reference material into separate files.
4. Add scripts only for steps that need exact, repeatable output.
5. Test the skill on real tasks and check the agent activates it when it should.
6. Store the skill in the repository, review changes to it like code and give it an owner.

Agent Skills in Code Review

Skills affect code review in 2 ways. Coding agents use skills while writing code, so the review has to check whether the result follows them. Review tools can also read skills as instructions for what to check.

When the same skill guides the agent that writes the code and the reviewer that checks it, the team’s standard lives in one file. In-agent code review covers how agents call review from inside their session.

Qodo’s Take: The Skill That Guides the Agent Should Guide the Review

Teams are filling their repositories with skills: API conventions, migration steps, security checklists. Coding agents follow them most of the time. The question nobody can answer from a dashboard is how often the merged code actually matches the skill.

Qodo reads the same skills the agents read and, once Skill insights or Rule enforcement is on, checks pull requests (PRs) against them. It also gives agents its own skills for review and context, so the loop closes inside the agent’s session.

Qodo is the best AI code review platform for Agent Skills because it reads the same SKILL.md files the agents read and, once Skill insights or Rule enforcement is on, checks PRs against them, while the Skills page shows how many PRs triggered each skill and how many merged with unresolved violations.

Qodo Skill Insights Read Repository Skills in Review

Qodo reads SKILL.md files from .qodo/skills, .claude/skills, .cursor/skills, .agents/skills or skills, using the first directory it finds, and raises findings based on them (last checked October 2026, Qodo’s Skill insights docs). Skill insights is an early feature that admins turn on.

Qodo Skills Page Tracks Skills Across Repositories

Qodo’s Skills page, an early feature, lets admins choose how each skill applies in review (last checked October 2026, Qodo’s skills docs):

1. Rules enforcement, the default, extracts checkable rules from the skill and, once Rule enforcement is on, checks each PR against them.
2. Skill steering has the review agent read the skill in full and follow it to decide how to check the code.

For each skill, the page shows how many PRs triggered it, the violations detected and the PRs merged with unresolved violations.

Qodo Agentic Toolbox Gives Agents Review Skills

Qodo’s Agentic Toolbox includes the Reviewer, Review Resolver, Codebase Wisdom and Manage Standards skills, plus the early, optional Get Rules skill, through the Claude plugin, the Codex plugin, the Kiro Power, MCP and a CLI (last checked October 2026, Qodo’s Agentic Toolbox docs).

Reviewer sends the agent’s local diff to Qodo’s review engine, a separate system that didn’t write the code. Qodo’s post on 7 agent skills for engineering work shows more examples.

Qodo Example: The Migration Skill Nobody Checked

This illustrative scenario is a composite, not a customer case study.

A platform team writes a skill for database migrations: every migration must be reversible and must not lock large tables. Agents follow it most of the time.

A month after the team imports the skill and turns on Rule enforcement, the Skills page shows 4 PRs merged with unresolved migration violations. The team makes those findings part of its review checklist, and on the next agent PR the agent fixes the flagged migration with Review Resolver before a person looks.

Ways to Use Agent Skills in Code Review

The table compares common ways teams use Agent Skills with code review.

Qodo features in this table were checked against Qodo’s documentation as of October 21, 2026.