Agent
Claude
Anthropic's reasoning model, wrapped in Wing's policy and audit envelope.
agent · governed
fabric · synchronized
Capabilities
What Claude does inside Wing.
CAP.01
Reasoned drafting
Long-context document and policy drafting with citation enforcement.
CAP.02
Tool-call routing
Wing-mediated tool calls with allowlists and per-tool rate limits.
CAP.03
Redaction at the edge
PII / secret scrubbing on every outbound prompt.
CAP.04
Reviewer handoff
Auto-escalates ambiguous decisions to a human approver.
Integration map
Where Claude reads and writes.
agent
Claude
↔Slack
←Notion
→Jira
←GitHub
←Okta
←Snowflake
← read→ write↔ read/write
Security & governance
guardrails
- Scoped OAuth tokens: never long-lived keys.
- Per-action allowlist with hash-pinned tool versions.
- HITL trigger on outbound to external domains.
- Token + prompt budget caps per tenant.
audit.log
00:42:11 claude.invoke ok tool=jira.create approver=auto 00:42:14 claude.invoke ok tool=slack.post approver=auto 00:43:02 claude.invoke HOLD tool=email.send approver=human 00:43:18 claude.invoke ok tool=email.send approver=u.j.lee