v2.0.0 Release: Plan/Design workflow, working memory fix, learning system review

[dean0x]

Overview

Final workstream to reach v2.0.0. Eight tracks of work (5 done, 3 remaining).

---

Track 1: Rename Shepherd → Evaluator ✅

• Rename shared/agents/shepherd.md → shared/agents/evaluator.md
• Update all references across skills, commands, agents, and orchestration files
• Update plugin.json manifests, tests, README, CLAUDE.md, CONTRIBUTING.md

Completed in PR #171

Track 2: Tester Agent + QA Skill ✅

• Create shared/skills/qa/SKILL.md + shared/agents/tester.md
• Add Tester to quality gate pipeline, /implement command, plugin.json manifests
• Add tests

Completed in PR #171

Track 3: Ambient Mode Refinements ✅

• Full pipeline rearchitecture: lean SessionStart preamble, SKILL router, three-tier depth classification
• All intents verified (REVIEW, RESOLVE, DEBUG), edge cases tested

Completed in PRs #172, #174

Track 4: Plan/Design Workflow + Gap Analysis ✅

Unified /plan command with gap analysis, design review, and plan-aware /implement. Implementation differed from original spec — unified into devflow-plan plugin with a designer agent rather than separate gap-checker/design-reviewer agents.

• Create gap-analysis and design-review skills + references
• Create shared/agents/designer.md (unified gap-checker + design-reviewer role)
• Extend shared/agents/synthesizer.md (planning mode)
• Create devflow-plan plugin (plugin.json, commands/plan.md, README)
• Add plan detection to /implement command (accepts plan documents)
• Deprecate /specify in favor of unified /plan
• Update src/cli/plugins.ts, marketplace.json, CLAUDE.md, README.md
• Build & verify (71 skills distributed, 614/614 tests passing)

Completed in PR #176

Track 5: Working Memory Message Extraction Bug 🐛

See #179 for details.

• Fix background-memory-update message extraction: tail -1 → head -1 for first user message
• Verify assistant extraction is correct (last message)
• Test with real session transcript

Track 6: Self-Learning System Review ✅ (scoped down)

Core learning system reviewed and stabilized in PR #173 (thresholds, single toggle, frontmatter, --reset). Remaining items (PF-004 god script decomposition, PF-006 jq latency, PF-005 hook duplication) are tracked in #23 (Tech Debt Backlog) — not v2 blockers.

• Fix self-learning thresholds, single toggle, frontmatter (fix: self-learning system — thresholds, single toggle, frontmatter, --reset #173)
• Review confidence scoring and temporal decay for correctness
• Review observation deduplication reliability
• PF-004 god script decomposition → deferred to tech debt
• PF-006 jq loop latency → deferred to tech debt
• PF-005 hook interface duplication → deferred to tech debt

Track 7: Project Knowledge Refinements 🔬

See #177 for details. Revisit the project knowledge system from scratch — analyze current decisions/pitfalls implementation, resolve orphaned PROJECT-PATTERNS.md, decide on two-file vs three-file model, evaluate extraction mechanisms and cross-workflow flow.

• Analyze current implementation end-to-end
• Decide file model and resolve PROJECT-PATTERNS.md
• Implement refinements
• Verify cross-workflow knowledge flow

Track 8: Ambient Mode Detection Tweaks 🎯

See #178 for details. Refine the lean classification approach based on real usage — intent accuracy, depth calibration, edge cases, router mappings.

• Collect observations from real usage
• Identify and fix misclassification patterns
• Apply classification/router adjustments

---

Release Checklist

• Tracks 1-4 complete
• Track 5 complete (fix: working memory extraction — first user message instead of last #179 — memory extraction bug)
• Track 6 scoped and core fixes shipped
• Track 7 complete (feat: revisit project knowledge system — analysis and refinements for v2 #177 — project knowledge refinements)
• Track 8 complete (feat: ambient mode detection tweaks — refine lean classification approach #178 — ambient detection tweaks)
• npm run build clean
• npm test — all tests pass
• Version bump to 2.0.0 in package.json (already done)
• CHANGELOG.md updated
• Tag v2.0.0 and create GitHub release