All notable changes to the PRD-Led Context Engineering template are documented in this file.
Format follows Keep a Changelog. Template version tracked in .claude/VERSION.
.claude/rules/directory with 6 modular rule files (01-session-protocols through 06-cross-agent-communication) — auto-loaded by Claude Code viaalwaysApply: truefrontmatterSoT/SoT.LESSONS_LEARNED.md— new SoT file with LL-XXX prefix for cross-session behavioral corrections and validated patterns- Staleness protocol in
SoT/SoT.UNIQUE_ID_SYSTEM.md— <30d current, 30-90d review, >90d verify before use MEMORY_ARCHIVE.mdfor all 4 agents — stores entries promoted to SoT during Phase E harvest.claude/hooks/traceability-gate.sh— PreToolUse hook (Write|Edit) verifies active EPIC before source code writes.claude/hooks/sot-sync-reminder.sh— PostToolUse hook (Write|Edit) reminds agent to update SoT after code writes- EPIC template: coordinator pattern (multi-agent routing table, synthesis checkpoint, self-contained worker prompts)
- EPIC template: memory harvest step in Phase E (agent memory → SoT.LESSONS_LEARNED.md)
- EPIC template: session ownership/locking with Active Session field (2-hour staleness threshold)
- Squad Status dashboard in README.md (agent activity + EPIC status tables)
- Cross-agent communication protocol rule (file-based async via SoT cross-references, EPIC observations, session state blockers)
- Git-based memory sync:
subagent-memory-save.shauto-stages MEMORY.md changes; commit conventionmemory({agent}): {summary} contextandallowed-toolsfrontmatter fields added to all 33 skillsCLAUDE.local.mdadded to.gitignorefor per-developer overrides- Context loading order documented as intentional cache architecture (stable→volatile for prompt cache efficiency)
- Agent renamed:
werk→devlabacross all files (AGENT.md, domain-profile.yaml, CHANGELOG, MIGRATION, all cross-references in horizon/studio/metro AGENT.md files) - Agent MEMORY.md simplified from 5-8 complex tables → 4 sections (Feedback, Patterns, Decisions, Handoff Notes) — prior format was never populated
subagent-memory-save.shupgraded from passivesystemMessage→ activehookSpecificOutputwithadditionalContext(matches load hook pattern)- CLAUDE.md slimmed from 114 lines → ~30 lines (pointer to
.claude/rules/) settings.jsonexpanded from 5 hook events → 7 (added PreToolUse, PostToolUse)- HOOK_CONTRACT.md updated with new hooks inventory, git commit conventions, Python dependency note
domain-profile.yaml— added LL prefix, agent registry updated (werk → devlab)
- Agent
werkremoved — renamed todevlab. Repos referencingwerkin EPIC leads, agent routing, or custom hooks must update todevlab - CLAUDE.md content moved — rules now in
.claude/rules/*.md. CLAUDE.md is a pointer only. Repos that grep CLAUDE.md for rule content must update to search.claude/rules/ - MEMORY.md format changed — old tables replaced with 4-section format. Repos with populated MEMORY.md files should migrate content into the new sections
- settings.json schema expanded — PreToolUse and PostToolUse entries added. Repos using
/localizemust re-sync settings.json
prd-v04-visual-prototype-gateskill — converts SCR- entries into structured prompts for Google Stitch (or equivalent UI generation tool), captures stakeholder feedback routed back to SoT IDs. Sits between Screen Flow Definition and v0.5 Red Team Review.prd-v06-environment-setupregistered in domain-profile.yaml, skills-inventory.md, and DEVLAB agent (skill existed on disk but was missing from all registries)
.claude/domain-profile.yaml— domain skills list updated (24 → 26 entries).claude/skills/skills-inventory.md— added Visual Prototype Gate and Environment Setup entries (Quick Nav, stage overview tables, full skill specs).claude/agents/studio/AGENT.md— addedprd-v04-visual-prototype-gateto Skills I Invoke.claude/agents/devlab/AGENT.md— addedprd-v06-environment-setupto Skills I Invoke.claude/README.md— updated v0.4 skill count (3 → 4) and v0.6 skill count (2 → 3)- Skill count: 26 → 28 PRD lifecycle skills + 5 methodology skills
.claude/VERSION— single source of truth for template versionCHANGELOG.md— structural changes documented per versionMIGRATION.md— step-by-step migration guide for derivative repos<!-- SECTION: -->and<!-- CUSTOMIZABLE: -->markers in all template files for merge-friendly syncing.claude/hooks/HOOK_CONTRACT.md— universal hook interface specification (language-agnostic).claude/hooks/context-validation.sh— shell variant of SessionStart hook.claude/hooks/context-density-gate.sh— shell variant of UserPromptSubmit hook.claude/hooks/sot-update-trigger.sh— shell variant of Stop hook.claude/domain-profile.yaml— parameterized ID prefixes, skill taxonomy (core vs domain), agent registry- Dependency manifests in all hook
.mddocumentation files template_versionfrontmatter field in CLAUDE.md, PRD.md, README_template.md, EPIC_TEMPLATE.md, SoT.README.md
- EPIC template sections: numbered (
## 0. Session State) → semantic (## Session State) - CLAUDE.md: "Section 0" references → "Session State" semantic references
- ID Ownership in CLAUDE.md now references
domain-profile.yamlas canonical registry .claude/README.mdupdated to reflect actual directory structure (agents, hooks, skills)- Agent files restructured: flat
.claude/agents/DEVLAB.md→.claude/agents/devlab/AGENT.md+MEMORY.md
- Agent paths changed:
agents/{NAME}.md→agents/{name}/AGENT.md+agents/{name}/MEMORY.md - EPIC section headers no longer numbered — scripts referencing "Section 0" must update to "Session State"
- See
MIGRATION.mdfor complete derivative update checklist
- Complete skill library: 26 PRD lifecycle skills (v0.1–v0.9) + 5 methodology skills (ghm-*) (note: count updated to 28 in v3.1.0)
- Skill-per-directory structure with SKILL.md + assets/ + references/
- 3 hooks: context-validation, context-density-gate, sot-update-trigger
- Hook .md documentation paired with each script
- 4 agent definitions: HORIZON, STUDIO, DEVLAB, METRO
- 12 SoT files standardized to ~100-150 lines each
- EPIC template with 5-phase execution model
- Progressive documentation protocol in CLAUDE.md
- README_template.md for derivative repos
- Standardized SoT files to consistent structure with YAML frontmatter
- Methodology renamed to "PRD Led Context Engineering"
- Initial template structure
- CLAUDE.md agent operating guide
- PRD.md progressive requirements template
- SoT directory with ID-based knowledge graph
- EPIC template for execution planning
- README.md methodology explainer