v0.8.0 — Shrink to MVP

Full Changelog: v0.7.1...v0.8.0

v0.7.1 — Harden CI Self-Test & Seed Detection

Full Changelog: v0.7.0...v0.7.1

v0.7.0 — Protocol Alignment & Retrieval Consistency

Full Changelog: v0.6.1...v0.7.0

v0.6.1 — Fix Route Leakage & Align Docs with v2.1

Full Changelog: v0.6.0...v0.6.1

v0.6.0 — Real Runtime Hook + Case Provenance

v0.6.0 — Real Runtime Hook + Case Provenance

P0: Case Provenance & Verification

• build_index.py now carries verified, source, is_seed into index entries
• Index exposes verified_case_count and synthetic_case_count
• Synthetic seeds isolated to cases/seeds/ directory — no longer mixed with real cases
• --exclude-seeds flag added to both build_index.py and retrieve_cases.py
• Case verification lifecycle defined: unverified → applied → verified (requires ≥2 resolved outcomes)
• resolutions array added to schema (legacy resolution field deprecated)
• record_outcome.py script to record outcomes and auto-verify cases

P1: Real Runtime Hook

• hooks/claude_code/post_tool_error.py — automated Claude Code PostToolUse hook
• Activation conditions (any one triggers):
  • Same tool failed ≥2 times in last 3 minutes
  • Tool response contains error/failed/exception patterns
  • User message matches rejection pattern (^(不对|错了|retry|再来|wrong|no|nope)\b)
• Behavior: builds intake card from context, retrieves similar cases, injects into agent context
• Falls back to seeds with low-confidence warning if no real matches found
• .claude/hooks/hooks.json — project-level hook configuration (copy-paste ready)
• Setup docs: hooks/claude_code/README.md

Full Changelog: v0.5.0...v0.6.0

v0.5.0 — Protocol + Deterministic Retrieval

v0.5.0 — Stuck-State Protocol + Deterministic Retrieval

Major convergence to protocol + runtime positioning:

• JSON-only contribution path — human fallback / form-to-JSON assembly removed
• jsonschema-based validation — cross-file rule consistency checks
• rules/*.yaml as single source of truth — schema enums auto-synced via generate_schema.py
• problem_family converged to 7 canonical categories — deprecated values auto-mapped with warning
• routes.yaml — added registry_version and deprecated_routes mechanism
• task enum canonicalized — no aliases, no fragmentation
• retrieve_cases.py — deterministic top-k case retrieval (AgentRX provides retrieval, not route recommendation)
• new_case_id.py — canonical case ID generation (YYYY-MM-DD-task-shortuuid8)
• All golden cases marked as synthetic-seed — resolution cleared, verified=false
• First real hook — tool_error_intake_prefill.sh
• SKILL activation tightened — only concrete failure signals, no generic uncertainty

Full Changelog: v0.4.0...v0.5.0

v0.4.0 — AI-Native Runtime & Closed Submission Loop

What's Changed

🔒 Architectural Closure

• Route IDs are now the single source of truth across all layers
• Removed legacy mapping layers between taxonomy and schema
• Strict JSON Schema validation with jsonschema library
• End-to-end contract tests for form → parse → validate pipeline

🔄 Closed Contribution Loop

• Automated ingestion pipeline: Issue → Parse → Validate → Save → Rebuild Index
• Workflow now auto-ingests accepted cases into the case library
• Task normalization layer with alias mapping for consistent retrieval

Vocabulary Cleanup

• Complete removal of legacy field names (skill_triggered, failure_type, observed_symptom, etc.)
• All documentation, hooks, and examples unified to v2.1 terminology
• CI lint rules prevent regression to old vocabulary

📊 Data Quality

• Unified UTC timestamps across index and cases
• Strict validation prevents schema drift
• Contract tests ensure form/parser/schema alignment

Full Changelog: v0.3.0...v0.4.0

Full Changelog: v0.3.0...v0.4.0

v0.3.0 — v2.1 AI-Native Release

Full Changelog: v0.2.0...v0.3.0

v0.2.0 — Renamed to AgentRX

Full Changelog: https://github.com/LpcPaul/AgentRX/commits/v0.2.0

v0.1.0 - Initial Release

Full Changelog: https://github.com/LpcPaul/skill-doctor/commits/v0.1.0