7 added, 68 removed. Audit A to A.
# CLAUDE.md
- > **SCOPE:** Entry point with rules and navigation ONLY. Guides in `docs/`. Workflows in `SKILL.md`. Public docs in `README.md`.
-
- Skills collection for Claude Code with config-driven Agile task management (Linear or File Mode).
-
- ## Critical Rules
-
- | Rule | Details |
- |------|---------|
- | **Architecture Guide** | Read `docs/architecture/SKILL_ARCHITECTURE_GUIDE.md` before any skill work |
- | **MANDATORY READ** | Use `**MANDATORY READ:** Load {file}`. Passive refs are NOT followed |
- | **Path Resolution** | Relative to skills repo root, NOT target project |
- | **Sequential Numbering** | 1, 2, 3 (NOT 1.5). Sub-steps: Na/Nb (3a, 4a) |
- | **Docs in English** | Stories/Tasks can be EN/RU |
- | **No version auto-updates** | Update ONLY when user explicitly requests |
- | **YAML quoting** | Wrap `description:` in quotes if it contains `:` |
- | **Research-to-Action Gate** | No defect = informational, not actionable |
- | **No hardcoded counts** | Counts ONLY in README.md badge |
- | **No Changes sections** | `**Version:** X.Y.Z` + `**Last Updated:**` at end |
- | **DoD with checkboxes** | `## Definition of Done` with `- [ ]` items |
- | **Worker independence** | No parent/peer refs in L3 Workers |
-
- ## MCP Tool Preferences
-
- Prefer `hex-line` for code/config/script/test files.
- - Use `outline` before large reads, then `read_file` with ranges.
- - Use `edit_file` / `write_file` for writes, `bulk_replace` for multi-file text rename, `verify` after conflicts or delayed follow-up edits, and `changes` for diff review.
- - Use `hex-graph` only for symbol, reference, architecture, and semantic diff questions.
- - With the hook active, project-scoped text `Read`/`Edit`/`Write`/`Grep`/`Glob` are redirected to `hex-line`. Built-in tools stay fine for images, PDFs, notebooks, and text paths outside the current project root.
-
- ## Quick Understanding
-
- | What | How |
- |------|-----|
- | Project overview + tree | `cat README.md` |
- | Architecture (L0-L3) | `cat docs/architecture/SKILL_ARCHITECTURE_GUIDE.md` |
- | Key workflow | `ln-700 -> ln-100 -> ln-200 -> ln-1000` |
- | Tool config (Linear/File) | `cat skills-catalog/shared/references/environment_state_contract.md` |
- | Skill metadata | `head -20 {ln-NNN}/SKILL.md` |
-
- ## Navigation
-
- **DAG:** CLAUDE.md -> `docs/README.md` -> topic docs. Read SCOPE tag first.
-
- | Topic | File |
- |-------|------|
- | Writing Guidelines | `docs/architecture/SKILL_ARCHITECTURE_GUIDE.md` section Writing Guidelines |
- | Environment State | `skills-catalog/shared/references/environment_state_contract.md` |
- | Risk-Based Testing | `skills-catalog/shared/references/risk_based_testing_guide.md` |
- | Frontmatter fields | `skills-catalog/shared/references/frontmatter_reference.md` |
- | Hooks reference | `skills-catalog/shared/references/hooks_reference.md` |
- | Questions format | `skills-catalog/shared/references/questions_format.md` |
- | Hook Design | `docs/best-practice/HOOK_DESIGN_GUIDE.md` |
- | MCP Tool Design | `docs/best-practice/MCP_TOOL_DESIGN_GUIDE.md` |
- | Token Efficiency | `docs/standards/TOKEN_EFFICIENCY_PATTERNS.md` |
- | Prompt Caching | `docs/best-practice/PROMPT_CACHING_GUIDE.md` |
- | npm Packages | `docs/standards/NPM_PACKAGE_BEST_PRACTICES.md` |
-
- ## Maintenance
-
- Version update (ONLY on explicit request): update `**Version:**` in SKILL.md, version in README.md tables, CHANGELOG.md paragraph.
+ <!-- Projection of AGENTS.md via the @ import. AGENTS.md is the canonical source. Do not duplicate content here — add it to AGENTS.md instead, or scope it to `.claude/rules/*.md` with a `paths:` filter. -->
- ## Compact Instructions
+ @AGENTS.md
- Preserve in priority order during /compact:
- - Architecture decisions and rationale (NEVER summarize)
- - Modified files and their key changes
- - Current verification status (pass/fail)
- - Open TODOs and rollback notes
- - Tool outputs (can delete, keep summary only)
+ ## Claude Code
- **Last Updated:** 2026-03-20
+ - `/compact` preservation order: architecture decisions, modified files, verification status, open TODOs, tool outputs as summaries only.
+ - Auto memory is on by default. Claude writes learnings to `~/.claude/projects/<project>/memory/` — run `/memory` to inspect or edit.
+ - Scope path-specific rules to `.claude/rules/*.md` with a `paths:` frontmatter filter rather than inlining conditional "when working on X" blocks here.
+ - Nested `CLAUDE.md` files in subdirectories load on demand — prefer them for area-specific guidance over growing this root file.