v2.0.0 — Knowledge-Graph-Assisted Architecture Passes

Breaking

• Removed Idea-to-Architecture Mode. Pure raw-idea work routes to sibling skill idea-to-architecture-agent. If unavailable, agent states the limitation instead of running a built-in workflow.

Added

• Assess step in core flow (... Map -> Assess -> Document ...). Every finding requires evidence, impact, severity, confidence, and smallest safe correction.
• Knowledge-graph-assisted mapping and assessment (optional). When a codebase intelligence MCP is available, Step 5 uses dependency and caller queries to accelerate mapping, and Step 6 treats graph cycles, layer-crossing edges, fan-in outliers, and zero-consumer exports as signals feeding existing assessment dimensions — no new finding category. Graph data maps into existing evidence-strength taxonomy (Direct/Inferred + Verify first: Yes), severity is never pre-assigned, and file reading wins when graph and source disagree. Without such a tool, the flow is unchanged.
• rules/assessment-rules.md with assessment dimensions, finding shape, non-findings, and depth-by-pass-level guidance.
• templates/debt-register.md and example debt register in examples/basic-web-app/output/.
• Anti-patterns: Map Without Judgment and Unsupported Criticism.
• Rule 0: default to Scan Mode — promotion requires evidence, trigger, and risk.

Changed

• README stripped to MVP (242 to 97 lines, scannable layout, version v2.0.0).

Fixed

• inspection-rules.md rule 9: includes codebase graph queries.
• Description now grounds "blast radius analysis" at Step 5 caller lists.

v1.1.0 - Right-Sized Architecture Passes

Senior Architect Agent v1.1.0

Senior Architect Agent is an AI architecture skill for existing system mapping, software architecture documentation, architecture mapping, Mermaid diagrams, and AI agent handoff.

This release adds right-sized architecture pass control so agents can reduce model load without weakening architecture discipline.

Added

• Scan Mode for compact architecture notes on small, bounded, exploratory, or low-risk work.
• Focus Mode for scoped module, workflow, subsystem, or boundary mapping.
• Full Mode for whole-system architecture mapping, future-agent handoff, unclear ownership, 3+ interacting modules, persistence, integrations, payment, authentication, security, deployment, or major workflow changes.
• Inspection budget guidance for large repositories.
• Existing architecture context reuse before re-mapping stable areas.
• Compact Architecture Note output shape.

Preserved

• Evidence-first inspection.
• Confirmed, inferred, assumed, unknown, risks, decisions, and handoff separation.
• Markdown as the architecture documentation source of truth.
• Mermaid as editable diagram source.
• SVG only as a secondary visual artifact.