Changelog

All notable user-visible changes are documented here.

Unreleased

• Clarify deliverable users, intended decisions and necessary detail; use observable work units and early checks of costly assumptions without fixed time/progress checkpoints or extra approval rounds.

• Add an opt-in 49-candidate module catalog and a serial Ubuntu/Docker workflow with separate AI/Build services, immutable input/configuration identity, protected expectations, bounded execution, retained evidence and human-review-required completion. Include offline success/failure examples, explicit Codex/.NET overrides and unavailable DB/Windows capability checks; existing Skill and single-task Runner remain independent.

• Require complete v2 checkpoints before importing PASS; retain the canonical case expectation, normalize case IDs, reject future active verification dates and unsupported cross-session Excel cases, and preserve session roles/stop/cleanup in supported workbooks. v1 run-level metadata remains compatible.

• Prioritize individual conditions by changed program guarantees and consequential impact; preserve old-plan failures and scope amendments, reuse sufficient evidence and review preparation cost before expanding fixtures or automation.

• Centralize condition modeling, mandatory checks, test/automation adoption and human exit/budget decisions in the test strategy; reuse existing v1/v2 checkpoints without a new solver or required fields.

• Route lifecycle phases and Human Work Plan v2 details on demand; keep one task index and canonical definitions instead of repeated full guidance.

• Reduce repeated hook policy with an always-present safety kernel and full reinsertion on lifecycle/state/target/policy changes.

• Add canonical XML/plan JSON context projection; FMW HTML remains human-only output, never a model/tool data source.

• Add optional Human Work Plan v2 with versioned procedures, selective trials, dependency/restart checks, complete condition mappings and measured effort references; render shared operations with independent checkpoints in briefs and functional-test Excel while retaining v1 plans/results.

• Reframe the lifecycle as a risk-tailored, feature-sized specification/data/implementation loop followed by selective trials, formal verification, QA/human authorization, handoff and operational feedback, preserving existing phase IDs and schema contracts.

• Align Decisions guidance with general availability; add opt-in current-work phase candidates, observation-first OpenAI hooks, bounded content context and configurable transport timeout.

• Expand evidence-backed verification diagram selection with a purpose-driven 17-view catalog (including lineage, CRUD, decision tables, dependency, traceability, context and batch) without changing the Feature Map schema or fabricating diagrams.

• Add an optional A/B/C decision aid for legacy feature additions, with minimal Clean Architecture boundaries, a fictional C#/.NET + DB walkthrough and existing compatibility/recovery test links.

• Add cross-task Harness learning guidance and monthly read-only size/link/repetition audits, with scoped improvement proposals, periodic semantic calibration and no automatic Skill rewrites.

• Consolidate README and Skill entry points into task-based routes; clarify implemented optional tools, the coordination proposal, human-owned business expectations and local workbook requirements without changing runtime behavior.

• Exclude local-test-evidence and test-results directories at any depth from plugin ZIPs; verify archive contents with synthetic fixtures on Linux and Windows CI.

• Add a provider-neutral DecisionProvider boundary while preserving legacy Jev configuration.

• Implement the public-beta OpenAI Decisions API adapter with named predicate/choice normalization, default-off content controls, sanitized failures and legacy Jev compatibility.

• Add an optional single-task runner with fixed paths/checks, Linux OS-isolated evaluation, bounded retries, durable checkpoints, local candidate evidence and independent revision-bound human review.

• Add an offline repair/retry example and mandatory real-sandbox CI checks; keep HIGH/CRITICAL work human-led and defer queues, multiple agents and UI expansion.

• Add an OpenAI Agents API self-hosted PoC with multi-agent read delegation and a single canonical writer.

• Add bounded Programmatic Tool Calling guidance for read/query/reduction stages and stdio xquery-mcp integration.

• Define one-writer-per-worktree rules for concurrent agent work.

• Add an optional repository-coordination architecture for multi-agent asynchronous writes, dependency scheduling, conflict control, Git publication, and rebuildable XML projections.

• Define Knowledge, Coordination, and Projection planes without mixing runtime queue state into Feature Map XML.

• Update the pinned xquery-mcp dependency to 2.5.1.

• Add an MCP contract smoke test that launches the pinned server, lists required tools, and calls xml_validate_schema.

• Normalize xquery-mcp 2.5.1 QueryResult JSON for XPath/XQuery execution, including success, empty-sequence, and structured-error handling.

• Add a fully fictional Feature Map example with Mermaid diagrams and a PowerShell HTML rendering guide.

0.4.0

• Add Feature Map schema version 1.3.
• Add optional Mermaid diagrams in diagrams/diagram.
• Render Mermaid diagrams in the XSLT HTML view.
• Prefer a local Mermaid 12.0.0 runtime with pinned-CDN fallback.
• Add offline Mermaid setup scripts.
• Add OSS repository templates, CI/release workflows, and recommended GitHub settings.

0.3.0

• Add context sufficiency gate and UserPromptSubmit hook.
• Add operable Japanese guidance with natural-Japanese rendering principles.
• Add Feature Map 1.2 goal/exclude and open/item@type.

0.2.0

• Add Codex lifecycle hooks and optional Jev completion decisions.
• Add xquery-mcp validation tracking.

0.1.0

• Initial Source-first Feature Map workflow.