October 07, 2026
Changes on October 7, 2026
Complete change list
- Linked background context and ideation to canonical Problems: when approving problem proposals, associated background context chunks travel directly with the canonical Problem as first-class context evidence rather than remaining stranded in the feedback record.
- Problem Detail Accessibility & Feedback Triage: Dynamic
aria-describedbyhint on disabled verification action, and dropdown actions parity for unlinking or excluding problem context evidence with persistent focus/open styling. - Added new Ideation taxonomy type (
Sparklesicon ✨) across Signal Evidence chunks and problem proposals, capturing exploratory thoughts, brainstorming, and open discovery questions without mislabeling them as rigid feature requirements. - Direct customer quote integrity: direct quotes in quotation marks and under "Original feedback:" headings are strictly preserved verbatim in their original voice and language (e.g. German) without translation or summarization into English paraphrase.
- Problem Proposals & Radical Design Parity: associated background context items are integrated directly inside proposed problem cards beneath supporting citations under
--- Context · {count} ---, matching the exact visual layout of approved canonical problems and eliminating disconnected bottom containers. Outer muted boxes and nested item borders are eliminated in favor of clean whitespace and radical design rhythm. - Signal evidence extraction splits structured dogfooding notes and community chat quotes into distinct, bite-sized evidence items, stripping artificial template headers ("User need:", "Proposed direction:", "Disposition:") to preserve authentic verbatim quotes.
- Problem Detail & Review Views: context evidence is visually and technically separated from supporting evidence. Context chunks are positioned at the bottom beneath a clean divider ('--- Context · {count} ---') without resolution checkboxes or redundant status badges. The Evidence section header, source reach count, and resolution verification evaluate only genuine supporting evidence, ensuring reference context never requires resolution or blocks initiative closure.
- Capped exploratory and ideation evidence severity to
S3(Low) to preserve discovery room and prevent premature escalation. - Eliminated ghost/blank evidence rows in problem proposal reviews when citing deleted or unlinked chunk references.
- Connected clients can link Work Items to existing Initiatives, Capabilities and Changes, distinguish direct links from inherited relationships, and locate Changes by repository or source ID. PR references link imported Issues in either arrival order; removing an automatic reference preserves manual links.
- Connected clients can correct Feedback title, summary and classification, close or reopen Feedback, and record explicit Change/evidence claims. Manual classifications survive automatic analysis. Closing affects only explicitly selected open Evidence and never closes a Problem or Initiative; edited evidence marks previous fix claims stale. Additional OAuth approval is required for the new write scopes.
- GitHub Work Items now follow the Issue's confirmed status. A merged PR or a reference alone no longer completes an open Issue; reopening clears its closed date. Explicit title, description and lifecycle edits remain pending until GitHub confirms them, with conflicts and failures visible.
- Background Intelligence Upgrades: Migrated theme synthesis, capability matching, duplicate/merge detection, and inbound feedback processing to native Structured Outputs, allowing deep reasoning without tool-invocation errors.
- Initiative Goal Suggestions with Reasoning: Fixed an issue where clicking "Suggest goal directions" in the Initiative workspace could fail with an API error ("Function tools with reasoning_effort are not supported"). Goal directions now generate reliably with high-accuracy reasoning models (
gpt-6-luna). - Connected MCP clients receive Capability-creation status based on their actual permission; old connections must approve additional access. Invalid Capability maturity values now list
concept,beta,stable, andoptimized. - Authorized Admin/Member MCP connections can edit existing Capability descriptions and other supported fields with version checks, create Problems from real evidence, and create draft Campaigns. These writes require their explicit grants and enabled server flags; Problem writes default off. Campaign creation sends no messages and exact retries return the same draft.
- Authorized MCP connections can explicitly link Capabilities to Initiatives and connect Feedback through selected Evidence and an existing Problem to an Initiative. Creation alone does not imply automatic linking; external proposal creation remains disabled.
- Connected external clients can create new Capabilities directly with a separately approved create permission for Admin/Member users. External proposal creation remains disabled. Implemented; the gateway rollout flag defaults off pending coordinated app/database deployment, and release is not verified.
- Agent Connections and MCP authorization no longer require enabling a Testlab option. Existing Settings role visibility and OAuth access checks still apply; simulation controls are development-only.
- Gateway v3 returns machine-readable tool failures in _meta.error instead of structuredContent.error, while preserving isError, readable messages and successful output schemas. Schema-validating clients can receive expected failures after tool discovery.
- Connected MCP clients can search Capabilities and list their children or relationships without rejecting successful responses because of extra pagination fields from the app.