v1.5.0 — Command Flow Map + Documentation Sync

What's New

Command Flow Map (new section in README)

• Full lifecycle mermaid diagram: `/start` → `/create-docs` → `/plan-sprint` → `/implement` → `/review-impact` → `/end`
• Command Reference table with internal execution paths, agents invoked, and output per command
• Command Pairing Examples for 4 common workflows

Reference Improvements

• Model names simplified across agent frontmatter: `claude-sonnet-4-6` → `sonnet`, `claude-opus-4-6` → `opus`
• `CLAUDE_AUTOCOMPACT_PCT_OVERRIDE` lowered from 80 → 60 (compact even earlier)
• Added `read.deny` rules for sensitive files: `.env`, `.pem`, `.key`, `secrets/`, `.aws/`, `.ssh/`
• Added `deny` rules for destructive commands: `rm -rf` and `git push --force`

Documentation Fixes

• Slash command count corrected: 13 → 12 (`plan.md` removed in v1.3.0)
• README changelog now includes v1.3.0 and v1.4.0 entries

---

Previous: v1.4.0 — Sprint Token Budget Standard + Per-Task Story Folders

feat: v1.4.0 — sprint Token Budget standard + per-task story folders

- plan-sprint.md: Phase 4 creates story.md per task with ## Token Budget
  section; Phase 5 saves STATE.md (was Phase 4 before)
- Effort in task tables = S/M/L only (token signal). Days only in Gantt.
- S/M/L rules: S → max 3-6 files, skip brainstorm, direct implementation
  M → max 4-6 files, one planning step | L → 8+ files, full cycle
- Phase 3 orchestrator prompt enforces S/M/L in task table
- seq.mermaid created for all flow-based tasks (auth, CRUD, API, CI)
- SKILL.md checklist: 3 new items for S/M/L + Token Budget enforcement
- diagrams-and-commands.md: /plan-sprint rewritten + 2 new design principles

feat: v1.3.0 — parallel agent execution, hooks, and statusline trace

Commands now wire agents via Agent tool (subagent_type + run_in_background).
Adds PreToolUse hook for active file tracking. Statusline upgraded with
agent name, turn counter, and active file segment.

v1.1.0 — Best Practices Update

What's New

Based on comprehensive analysis of claude-code-best-practice (84 tips, 9 reports, hooks docs, skill/agent implementations).

Agent Improvements

• All 13 agents now have full frontmatter: maxTurns, color, memory, skills preloading, permissionMode
• Dev agents (01-05) use permissionMode: acceptEdits — auto-accept file edits
• Dev agents preload quality skills automatically (architect→design-patterns, devs→refactoring-techniques, qa→code-quality, devops→deploy)

Skill Improvements

• Skills now use directory format: skills/<name>/SKILL.md (auto-discovered)
• effort: high on brainstorm and plan skills (deeper thinking)
• context: fork on brainstorm skill (isolated subagent context)
• Two skill patterns documented: Agent Skills (preloaded) vs Invocable Skills

New Files Generated

• settings.json — Permissions (git pull auto, push ask; gh read auto, write ask), status line, spinner tips, attribution disabled, autocompact at 80%
• statusline-command.sh — Colored status bar: model (magenta), tokens (cyan), folder (blue), git branch (green/yellow/red), context bar (adaptive)

CLAUDE.md Improvements

• Rule 11: "Verify your work" — 2-3x quality improvement
• <important> conditional tags for critical rules
• Orchestration Pattern: Command → Agent → Skill
• Workflow Best Practices section

Command Improvements

• model: haiku on lightweight commands (start, end) — saves tokens/cost
• argument-hint and disable-model-invocation on relevant commands

See CHANGELOG.md for full details.

v1.0.0

feat: initial release — 13 agents, 6 rules, 11 skills, 12 commands