fix(sdk): clarify skill source labels in system prompt (#3464)

The skills section of the system prompt listed sources as `**User
Deepagents Skills**` and `**User Agents Skills**` with no explanation of
what that distinction means. Models reading the prompt had no basis for
understanding that "Deepagents" sources are scoped to this agent tool
while "Agents" sources are shared across all agent tools on the machine.

## Changes

- Adds a single explanatory sentence to `SKILLS_SYSTEM_PROMPT` in
`SkillsMiddleware`, inserted between the locations list and `**Available
Skills:**`, clarifying that "Deepagents" sources are agent-specific and
"Agents" sources are cross-tool shared
This commit is contained in:
Mason Daugherty
2026-05-18 22:23:15 -07:00
committed by GitHub
parent bfbb8bc557
commit fc6a24f188
2 changed files with 4 additions and 0 deletions
@@ -786,6 +786,8 @@ You have access to a skills library that provides specialized capabilities and d
{skills_locations}{skills_load_warnings}
Sources labeled "Deepagents" are specific to this agent tool; sources labeled "Agents" are shared across all agent tools on this machine.
**Available Skills:**
{skills_list}
@@ -63,6 +63,8 @@ You have access to a skills library that provides specialized capabilities and d
**User Skills**: `/skills/user/`
**Project Skills**: `/skills/project/` (higher priority)
Sources labeled "Deepagents" are specific to this agent tool; sources labeled "Agents" are shared across all agent tools on this machine.
**Available Skills:**
- **web-research**: Structured approach to conducting thorough web research on any topic