Browse specs
Skills and tools compiled to WASM. Immutable versions, SHA-256 verified, installable into any agent in your team.
Tenant context access (reads + full lifecycle writes: company + team + workspace + run state). Ships its tools via the TenantHandle (Pattern B, fourth domain handle after Collective/Governance/Commerce). All reads share the same KV-walk shape; the handle composes naturally as new tenant-scoped enumerations + aggregations + lifecycle mutations land. Every tool ships an output_view (Block Kit JSON in views/) per the SPEC architecture's UI-component point. The remaining mutation tools (update_tenant, suspend_agent, ...) stay in the kernel — they need richer auth + audit hooks the SDK doesn't yet model.
Stripe vocabulary — read account balance, list charges. The connected-account variant of @platform/commerce's checkout flow. Stage 1C — bundled 1 legacy skill(s) from data/skills/ so SPECs are the only skill source (agent.skills[] can be deprecated).
Jira vocabulary — search issues via JQL, read + create + comment + update via the v3 REST API.
Notion vocabulary — search the workspace, read pages. Authors against a connected Notion integration.
Slack messaging vocabulary — send messages and list channels through a connected Slack workspace.
Gmail vocabulary — send mail and list labels through a connected Google account.
GitHub vocabulary — repos, branches, commits, files, issues, PRs, releases. The full read+write surface against a connected GitHub account.
Research workflow — web search, source fetching, citation-disciplined synthesis.
Message-composition + structured reporting / approval + stakeholder notifications + A2A bridge.
Collective-intelligence primitives — peer ratings + lessons learned + per-agent retrospective + cross-prefix data query + topic-fanned insights + tenant-shared cognition. Ships 9 tools (rate_agent, log_lesson_learned, reflect, retrospective, query_data, publish_insight, subscribe_insights, shared_memory_write, shared_memory_search). First Pattern B SPEC: introduces CollectiveHandle in the SDK that exposes scoped reads/writes over agent_rating + lesson_learned + cost_report + collective_report + insight + insight_all KV prefixes + the tenant-shared MemoryStore namespace (__shared__). The richer collective tools (request_feedback, compare_approaches, request_review, propose_decision, cast_vote, ...) stay as kernel builtins — they need additional SDK surface that CollectiveHandle doesn't yet model.
Read-only discovery surface — browse integration types, skills, LLM models, AND the SPEC catalog via the PlatformAssetsHandle (Pattern E).
Read-only introspection of the agent's own configuration. Ships 7 tools: read_own_prompt, list_my_specs, list_my_skills, list_my_tools, read_own_model, list_my_permissions, check_own_budget. self_update_prompt remains as a kernel builtin (write path with CAS retry + AgentVersionManager).