CedarCopilot wants to merge 8 commits into staging from feat/mcp-coverage-buckets-1-3
Live on prod, no production signal yet
Verification reached its budget before a verdict.
Unable to complete verification: the available telemetry query results only matched unrelated PostHog `/flags/` spans, and no concrete `config-write:create-field` production execution or success result was identified. Verdict: inconclusive.
Unable to complete verification: telemetry returned no production spans matching config-write since the deploy, so there is no positive runtime evidence for either the explicit-null or omitted-owner behavior.
Verification reached its 120s wall-clock budget before a tied verdict. Libra is keeping this intent verifying until the next check.
Verification reached its 130s outer budget before the verifier returned. Libra is keeping this intent verifying until the next check.
Unable to verify: no matching `read-connection-settings` spans were found in production telemetry after the deployment, so there is no positive runtime evidence of successful execution or payload exposure.
Libra has verdicts on 0 of 18 tracked behaviors on prod; 18 are still being checked. Libra checks hourly for 3 days after each deploy.
Closes buckets 1-3 of the MCP coverage roadmap , 10 independent streams, each a logical commit group, cherry-picked onto one branch so they can be reviewed one at a time. Two items (create-field + CRM-picklist sourcing, connection-settings ergonomics) were the stated top priority; two others are the confirmed root-cause fixes for a real customer incident (Adapt Insurance / Karim, webhook testing).
Every stream was tested by calling its tool's execute() directly against a real Cedar account (the Cedar MCP connector is unreachable from this dev environment) , real creates, real psql before/after, real cleanup, not just green tool-call returns. /refresh-agent-knowledge was run against cedar-configuration wherever a stream touched agent-facing tool schemas/triggers.
options, one-step CRM-picklist sourcing + auto-link (sourceFromCrm), ownerAgentId, remove-field-option resolution step, set-stage-criteria action.stageRequiredFields (with ergonomic per-stage set/remove, not just whole-object replace), dealSyncFilter, customObjectConfig, fetchUnownedDeals, autoSyncExternalDms, notification-channel settings, CRM mapping type fields , read+write parity throughout.manage_playbook_webhook gains op:list/op:test. Direct fix for the Adapt Insurance incident (no customer-facing way to test-fire a webhook-triggered agent).sourceUrl and generic upload, alongside the existing Gmail-attachment source.ui-config gains create/update/delete-canvas, wrapping the existing tRPC logic.subagent_execution:user (SELF_OR_ORG_ADMIN) authorize() policy row.documents/agent_operating_procedures/agent_executions, same tenant-scoping rigor as every existing entry.account-setup-from-transcript (dead SQL, a manual step that's already a one-call action), plus webhook added to choosing-a-trigger.md's decision ladder.test-run-subagent.test.ts tests broke when combined (not visible in any single stream's isolated testing): they assumed authorize() was called at most once; the targetUserId refactor made it always call once up front for self/target resolution. Fixed the test mocks, no production code changed.pnpm --filter @zero/server run types) clean on the full consolidated branchconfigTool (130), field-deletion-safety, canvas-crud, guarded-write (37), sql-guard (+7 new), test-run-subagent (19, fixed), debugTool, full account-config directory (144), full auth directory (355)/refresh-agent-knowledge run wherever a stream touched agent-facing schemas , all clean or fixedš¤ Generated with Claude Code
https://claude.ai/code/session_01F4mjBtjYZyxnVvLQpehDSb
This PR substantially expands MCP administration coverage for fields, connection settings, webhooks, documents, canvases, SQL access, and playbook testing. Changes since the previous review refine several of those features:
Libra has not measured any production surfaces for this change yet.