Skills

Skills versus prompts and memory

Use the current prompt for this job's changing request. Use memory for durable facts, preferences, people, and decisions. Use a skill for the repeatable process and quality bar.

For example, a customer-research skill can define source requirements, interview synthesis steps, confidence labels, and the expected report structure without storing one project's findings in the skill.

Built-in skills

Pazi provides managed skills for core platform capabilities such as web work, browser permissions, memory, scheduled tasks, chat cards, image generation, agent control, and common file or service workflows.

Managed skills are maintained by Pazi, refreshed when the runtime changes, and hidden from the Skills page so they cannot be edited or deleted accidentally. Some are role-specific; the operating officer receives coordination skills ordinary specialists do not need.

Add a custom skill

Open All Docs → Skills for the selected agent and choose Add new. Give the skill a concise name, a description that says when it applies, and clear Markdown instructions.

You can also ask the agent to create a skill after a successful workflow. Review what it writes before relying on it repeatedly.

Structure useful instructions

Discovery and invocation

The agent sees available skill names and descriptions and loads the full instructions when a task matches. You usually do not need to name a skill explicitly, but doing so can remove ambiguity when several skills overlap.

Write descriptions that distinguish one skill from another. Avoid making every skill claim to apply to all work.

Manage skills

Select a custom skill to review its content. You can edit or delete user-created skills at any time. Changes apply to future runs and conversations that invoke the skill.

Skills belong to the selected agent's durable workspace. Pazi reconciles managed and custom skills into the active runtime without intentionally removing the agent's product-visible skills.

Keep skills safe and maintainable

Do not put passwords, API keys, tokens, private keys, or temporary login links in a skill. Use the secure credential flow.

Prefer a small number of focused, well-named skills over one enormous instruction file. Update a skill when the real workflow changes, and remove obsolete instructions that conflict with current practice.