SkillsCopy Anchor Link

Prompt skills are reusable instruction blocks. Action Agent Studio merges their content into the Agent's system prompt; it does not install executables, shell commands, or callable tools.

Organization and platform ownershipCopy Anchor Link

Action Agent Studio validates that every selected organization skill belongs to the Agent's organization and that every selected platform skill exists and is not deleted. The Agent stores references to both libraries, not copied content.

Prompt injection orderCopy Anchor Link

For each runtime configuration load, the Agent prompt is assembled in this fixed group order:

1. the Agent's trimmed base system prompt;
2. every attached, non-deleted platform skill content block; and
3. every attached, non-deleted organization skill content block.

Non-empty blocks are joined with one blank line between blocks. If the complete result is empty, the runtime uses You are a helpful assistant.

The current loader does not define a semantic sort inside the platform-skill group or the organization-skill group. Do not make correctness depend on one skill preceding another within the same group. Put ordering requirements in the base prompt or an explicitly attached organization skill.

Skills are appended to the base prompt; they do not replace it. Keep the base prompt concise and put reusable policy, tone, and product facts in skills. All blocks still count toward the selected model provider's context limit.

Skills do not grant permissions. When the organization has published input-stage guardrail rules, the current input policy inspects the trailing user message; it never inspects the base system prompt or injected platform/organization skill blocks, and an organization that has published no input-stage rule has nothing inspecting the message at all. Review and moderate skill content separately.

Attach and refreshCopy Anchor Link

1. Open the Skills panel in Action Agent Studio.
2. Attach zero or more organization skills and platform skills.
3. Save, then start the next message.
4. Ask the Agent to restate a critical policy when you need to verify the saved order and content.

Attaching or detaching skills is an Agent-owned write and increments configVersion. The runtime loads Agent configuration during several phases, so a later phase can observe a newer selection during the same in-flight turn; configVersion is not an immutable whole-turn snapshot.

The Agent configuration cache has a 60-second TTL. An Agent attachment change bypasses that stale cache through the version check, while an edit to the content of an already attached skill can take up to the TTL to converge if the edit does not increment that Agent's version.

Deleted skills are omitted. A skill with empty content after trimming is also omitted. Duplicate selected IDs are collapsed when Action Agent Studio saves the attachment.

What skills do not doCopy Anchor Link

• They do not grant organization, model-provider, MCP, or remote API permissions.
• They do not become HTTP or MCP tools.
• They do not run in the browser or ActionFlows server as code.
• They do not replace a provider's function-calling requirement for tools or MCP.

Organization skills can also attach to supported catalog AI nodes such as Generate Text. That is a separate Flow-node prompt assembly and does not create an Agent session or tool.

RelatedCopy Anchor Link

• Skills
• Action Agent lifecycle
• Model selection
• Knowledge