Agentic Workflow
One of the greatest strengths of agentic tools is work delegation. By leveraging the power of agents, you can write a sequence of instructions and let the agents execute them for you. This allows you to focus on the high-level design of your workflow, while the agents take care of the details.
For example, instead of manually fetching a Sentry crash, creating a Jira ticket, writing a fix, and opening a PR — you describe the outcome once and the agent handles every step:
You: "there's a crash in Sentry, GOOD-DOCTOR-ANDROID-6YW, fix it"
Agent: [fetches issue] → [identifies root cause] → [creates Jira ticket]
→ [writes fix] → [opens PR]
The key is giving the agent enough context about your team's process. That's what skills are for.
Skills
To let the agents find the right instructions to execute, you can create SKILL.md files that contain the instructions for a specific workflow. The agents will then use these instructions to execute the tasks you have assigned to them.
This concept draws from the Agent Skills open standard — a simple, open format for giving agents new capabilities and expertise. Skills are designed to be portable across agent products so you can write them once and use them across different tools.
To learn more about how to write effective skill instructions, see Writing Your First Skill.
For how to prompt and use skills in practice, see How to Prompt with Skola Skills.
Extracting skills from conversations
Skills can also be extracted through a collaborative process. Work together with the agent until it produces the expected outcome, then ask the agent to extract the process into a skill instruction — for example: "extract the process we just followed into a reusable skill".
To help write or extract a new skill, you can use the Skill Creator — a specialized skill designed to help users create new skills from their conversation history.
Where skills live
Project level
The project-level skill instruction is placed in the repository and shared with all team members. The directory structure is:
project-root/
└── .agents/
└── skills/
└── <skill-name>/
└── SKILL.md
Some tools may expect skills in a tool-specific directory instead of .agents/skills. If your agent cannot find the shared skills directory, point its tool-specific path at .agents/skills.
For example, in Claude Code you can create a symbolic link so .claude/skills refers to .agents/skills:
ln -s ../.agents/skills .claude/skills
User level
User-level skills are personal and available across all your projects. Each agentic client has its own directory:
| Tool | Path |
|---|---|
| Claude Code | ~/.claude/skills/<skill-name>/SKILL.md |
| Opencode | ~/.config/opencode/skills/<skill-name>/SKILL.md |
| Codex | ~/.agents/skills/<skill-name>/SKILL.md |
A different agentic client could also use another agent's directory if configured to do so. Note that some clients require a session restart to recognize newly installed skills.
The Skola Marketplace
The Skola Marketplace is a shared registry of skills built by teams across the company. Browse it before building from scratch — install a skill with a single command:
skola install skill/sentry-to-pr
skola install skill/fix-ci
To publish your own skill, see Writing Your First Skill and Publishing a Skill.
