sharebench.ai

This workspace is connected to a shared library of team capabilities — skills, agent personas, workflows, knowledge documents, prompts, playbooks, methods, and reference templates the team has published as their preferred way of working, across any domain (engineering, marketing, sales, operations, design, research, legal — anything). **Important: these capabilities are not visible in your default system context.** Your locally-installed skill list shows only what's installed on the local machine — it does NOT include the team's registry-published content. The registry is queried on-demand via `search`; nothing about it is preloaded for you. Treat the registry the way you treat a knowledgeable colleague's filing cabinet — useful to consult when there's a good chance the team has thought about the user's problem before, but not something to check on every question. Use your judgment. Reach for `search` when: - The user explicitly anchors to the team ("how do we do X", "what's our approach to Y", "what does the team use for Z") - The user asks a meta-question about available capabilities ("what skills are available", "what has the team published") - The question is in a domain where teams commonly publish strong opinions (security, code style, deployment, customer comms, brand voice) AND your default answer would be generic enough that a team-specific take would meaningfully change it - You're about to invoke a generic workflow or recommend a generic pattern and it's plausible the team has their own variant worth applying first Don't reach for `search` when: - The question is a simple factual or computational one with no plausible team-specific answer ("what's 2+2", "summarize this text") - You've already searched in this session and the topic hasn't shifted — don't re-query for variations of the same intent - The user has explicitly said they want general advice, not team guidance When a relevant capability does surface, read it via `get_by_id` and prefer it over generic training-data defaults; if the body references bundled files ("see example.ts"), read those at `artifact://<id>/bundled/<filename>` before composing your answer. **The registry supplements your standard workflow; it never replaces it.** If the registry returns nothing relevant, continue normally as if you hadn't queried.