xtrm-tools 0.5.0

package/CHANGELOG.md

@@ -0,0 +1,504 @@
+# Changelog
+
+All notable changes to Claude Code skills and configuration will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+---
+
+## [Unreleased]
+
+### Added
+
+- **`AGENTS.md` — bd (beads) issue tracking section**: Comprehensive reference for the `bd` CLI covering session protocol, issue creation, dependency management, search/view commands, advanced features (`agent`, `gate`, `mol`, `audit`), and a bd vs TodoWrite decision table.
+- **`xtrm install project all` / `xtrm install project '*'`**: Non-interactive install of every available project skill into the current repository.
+
+### Fixed
+
+- **Claude-only target detection**: `xtrm install all` now enumerates Claude Code targets only, instead of surfacing stale Gemini/Qwen/Agents paths.
+- **Project-skill install-all coverage**: Added regression tests to verify merged hook counts and copied assets across all shipped project skills.
+
+### Roadmap
+
+- `AGENTS.md` should be included in the `jaggers-config install-service-skills` installation flow so all target projects get agent context out of the box (see ROADMAP.md).
+
+---
+
+## [2.0.0] - 2026-03-12
+
+### Added
+
+#### Project Skills Engine
+- **`cli/src/commands/install-project.ts`**: Generic "Plug & Play" project skill installer with deep merge for `settings.json` hooks
+- **`cli/src/commands/help.ts`**: Self-documenting help command with full CLI reference
+- **Project skills directory structure**: `project-skills/<skill>/.claude/` standard for modular tool packages
+
+#### Project Skills (5 skills shipped)
+- **`service-skills-set`**: Docker service expertise with SessionStart, PreToolUse, PostToolUse hooks
+- **`tdd-guard`**: Test-Driven Development enforcement with PreToolUse, UserPromptSubmit, SessionStart hooks
+- **`ts-quality-gate`**: TypeScript/ESLint/Prettier quality gate with `quality-check.cjs` (ported from bartolli/claude-code-typescript-hooks)
+- **`py-quality-gate`**: Python ruff/mypy quality gate with `quality-check.py` (custom implementation)
+- **`main-guard`**: Git branch protection with `main-guard.cjs` (blocks direct edits to main/master)
+
+#### Installation Commands
+- **`xtrm install`**: Global installation (replaces `sync`)
+- **`xtrm install all` / `xtrm install '*'`**: Non-interactive global install across all known targets
+- **`~/.agents/skills`**: Skills-only target added so the installed `skills/` tree is available without touching hooks/config
+- **`xtrm install project all` / `xtrm install project '*'`**: Install every project-specific skill package into the current repository
+- **`xtrm install project <tool-name>`**: Install project-specific skill package
+- **`xtrm install project list`**: List available project skills with descriptions
+
+### Changed
+
+#### CLI Rebranding
+- **Package renamed**: `jaggers-agent-tools` → `xtrm-tools`
+- **Binary renamed**: `jaggers-config` → `xtrm`
+- **Version bumped**: 1.7.0 → 2.0.0 (breaking changes)
+
+#### Command Restructure
+- **`sync` command** → renamed to `install` with updated messaging
+- **Default action**: Now shows help instead of running sync automatically
+- **`add-optional` command**: Removed (optional MCP servers now part of `install`)
+
+#### Architecture Decision
+- **Claude Code only support**: Removed multi-agent hook translation for Gemini/Qwen
+- **Focus**: Robust, well-tested Claude Code installation engine
+
+### Removed
+
+#### Multi-Agent Support
+- **`cli/src/utils/transform-gemini.ts`**: Deleted (Gemini hook translation)
+- **`cli/src/adapters/gemini.ts`**: Deleted (Gemini adapter)
+- **`cli/src/adapters/qwen.ts`**: Deleted (Qwen adapter)
+- **`transformToGeminiHooks`**, **`transformToGeminiFormat`**: Removed from `config-adapter.ts`
+- **Gemini/Qwen command generation**: Removed from `sync-executor.ts`
+
+#### Deprecated Commands
+- **`jaggers-config add-optional`**: Superseded by `xtrm install`
+- **`jaggers-config sync`**: Superseded by `xtrm install`
+
+### Fixed
+
+- **Project skills structure**: Standardized `.claude/settings.json` + `.claude/skills/` format
+- **Hook paths**: Corrected `$CLAUDE_PROJECT_DIR` references in all project skills
+- **Documentation**: README.md updated with accurate skill list and installation instructions
+
+### Documentation
+
+- **README.md**: Added Project Skills section, manual setup guide for Gemini/Qwen users
+- **Updated installation instructions**: `npm install -g github:Jaggerxtrm/xtrm-tools` recommended
+- **Each project skill**: Includes `README.md` and `SKILL.md` with usage guide
+
+### Migration Guide
+
+#### For Existing Users
+
+```bash
+# Old command (no longer works)
+jaggers-config sync
+
+# New command
+xtrm install
+
+# Global installation (recommended)
+npm install -g github:Jaggerxtrm/xtrm-tools
+
+# One-time run
+npx -y github:Jaggerxtrm/xtrm-tools install
+```
+
+#### For Gemini/Qwen Users
+
+Automated hook translation is no longer supported. See README.md "Manual Setup for Gemini/Qwen" section for manual configuration instructions.
+
+---
+
+## [1.7.0] - 2026-02-25
+
+### Added
+
+#### GitNexus Integration
+- **Optional MCP server**: `gitnexus` added to `config/mcp_servers_optional.json` with auto-install support (`npm install -g gitnexus`)
+- **PreToolUse hook**: `hooks/gitnexus/gitnexus-hook.cjs` — enriches Grep/Glob/Bash tool calls with knowledge-graph context via `gitnexus augment`
+- **4 knowledge-graph skills**: `skills/gitnexus/{exploring,debugging,impact-analysis,refactoring}/SKILL.md` — synced via standard pipeline
+
+#### Unified 3-Phase Sync Flow
+- **`cli/src/core/preflight.ts`**: Parallel `Promise.all` preflight checks across all targets. Returns `PreflightPlan` with file diffs, MCP status, and optional server list. Per-target error isolation — one bad target never aborts the rest.
+- **`cli/src/core/interactive-plan.ts`**: Single `prompts` multiselect plan — all targets, files, MCP servers, and optional servers in one view. `[~]` drifted and `[?]` optional items pre-unchecked by default.
+
+#### MCP CLI Sync
+- **`sync-mcp-cli.ts`**: Unified MCP CLI sync for Claude, Gemini, and Qwen via official `mcp add/remove/list` commands. Idempotent — re-running is always safe.
+- **Env file management**: `~/.config/jaggers-agent-tools/.env` — auto-created on first sync, validates required env vars (e.g. `CONTEXT7_API_KEY`), preserves existing values.
+- **ConfigAdapter enhancements**: Qwen and Antigravity support added; `type` field auto-handled per agent; `EnvVarTransformer` extended for cross-agent compatibility.
+
+### Changed
+
+#### Sync Command — 3-Phase Rewrite
+- `cli/src/commands/sync.ts` fully rewritten: Phase 1 preflight spinner → Phase 2 multiselect plan → Phase 3 ordered execution (prerequisite installs → file sync → MCP sync → post-install messages)
+- `--dry-run`: displays full plan grouped by target, prints "Dry run — no changes written", exits cleanly
+- `-y`/`--yes`: auto-applies pre-checked defaults without prompting
+- `--prune`: propagated through `plan.syncMode` to `executeSync` correctly
+- `--backport`: reverses sync direction (local → repo)
+
+#### sync-executor.ts
+- Removed inline `promptOptionalServers` call and manifest-based prompt tracking
+- Added `selectedMcpServers?: string[]` parameter — optional server names pre-selected upstream in Phase 2
+
+#### MCP Configuration
+- Split into `config/mcp_servers.json` (core: serena, context7, github-grep, deepwiki) and `config/mcp_servers_optional.json` (optional: unitAI, omni-search-engine, gitnexus)
+- `_notes.install_cmd` and `_notes.post_install_message` metadata — drives Phase 3 auto-install
+- Core servers: removed unused `filesystem`, `git`, `memory`, `gmail`, `yfinance-market-intelligence`
+- `serena` command updated to uvx-from-git with auto project detection
+
+#### Exported Symbols
+- `getCurrentServers(agent)` and `AgentName` exported from `cli/src/utils/sync-mcp-cli.ts` (consumed by `preflight.ts`)
+
+### Deprecated
+- **`jaggers-config add-optional`**: now prints a redirect notice — optional servers are part of `jaggers-config sync`
+- **JSON file sync for Claude/Gemini/Qwen MCP**: superseded by official `mcp` CLI method
+- **Repo `.env` files**: use centralized `~/.config/jaggers-agent-tools/.env`
+
+### Removed
+- **Old Claude-specific sync**: `cli/lib/sync-claude-mcp.js` (replaced by unified `sync-mcp-cli.ts`)
+
+### Fixed
+- **`--prune` propagation**: `runPreflight` now sets `syncMode: 'prune'` when `--prune` passed (was hardcoded `'copy'`)
+- **Optional server "already installed" filter**: now uses live `getCurrentServers()` call per agent instead of only checking core MCP names
+
+### Documentation
+- Updated SSoT: `ssot_jaggers-agent-tools_installer_architecture` → v1.4.0
+- Updated SSoT: `ssot_cli_ux_improvements` → v2.0.0
+- Updated SSoT: `ssot_cli_universal_hub` → v2.2.0
+- Updated SSoT: `ssot_cli_mcp_servers` → v3.2.1
+
+---
+
+## [1.6.0] - 2026-02-24
+
+### Added
+
+#### Documenting Skill Hardening
+- **`drift_detector.py`**: New script with `scan`, `check`, and `hook` subcommands — detects stale memories by cross-referencing `tracks:` globs against git-modified files
+- **`tracks:` frontmatter field**: Each memory now declares which file globs it documents; added to schema, all templates, and all 11 existing memories
+- **Intra-memory INDEX blocks**: `validate_metadata.py` now auto-generates a `<!-- INDEX -->` TOC table inside each memory from `##` headings + first-sentence summaries — allows agents to navigate without reading full documents
+- **Stop hook**: `config/settings.json` wired with Stop hook → `drift_detector.py hook`; fires at session end, injects a one-line reminder only when stale memories detected (zero token cost when clean)
+- **23 tests**: `test_validate_metadata.py` (4) and `test_drift_detector.py` (8, including `**` glob regression tests) added to existing suite
+
+### Changed
+- **`validate_metadata.py`**: INDEX generation now unconditional (no longer blocked by schema validation errors)
+- **`SKILL.md` workflow**: Rewritten with drift-first 5-step protocol and decision table (new feature → SSOT, bug fix → changelog only, etc.)
+- **All 11 existing memories**: `tracks:` globs added; INDEX blocks regenerated
+
+### Fixed
+- `extract_headings`: closing ` ``` ` was captured as section summary due to `in_code` toggle firing before capture check — fixed with `continue`
+- `match_files_to_tracks`: `**/` expansion was producing `*.py` (too broad); replaced with recursive segment-by-segment `_match_glob` helper
+- `inject_index`: frontmatter split hardened with anchored regex to prevent corruption on non-standard file openings
+- `generate_index_table`: anchor generation collapsed consecutive hyphens from stripped `()/` chars
+
+### Documentation
+- Updated SSOT: `ssot_jaggers-agent-tools_documenting_workflow_2026-02-03` → v2.0.0
+
+---
+
+## [1.5.0] - 2026-02-23
+
+### Added
+
+#### Service Skills Set (`project-skills/service-skills-set/`)
+- **Complete rewrite** of project-specific service skill infrastructure — replaces deprecated `service-skill-builder`
+- **Trinity skills** installed into `.claude/skills/` of any target project:
+  - `creating-service-skills` — 3-phase workflow: scaffold → Serena LSP deep dive → hook registration
+  - `using-service-skills` — SessionStart catalog injection + PreToolUse skill enforcement
+  - `updating-service-skills` — PostToolUse drift detection
+- **Scripts**:
+  - `scaffolder.py` — generates SKILL.md skeleton, script stubs, and auto-detects official docs from 30+ technology mappings (Docker images, requirements.txt, Cargo.toml, package.json)
+  - `deep_dive.py` — prints Serena LSP-driven research protocol with tool table for Phase 2
+  - `cataloger.py` — SessionStart hook; outputs ~150-token XML service catalog
+  - `skill_activator.py` — PreToolUse hook; territory glob + Bash command matching; injects skill load enforcement
+  - `drift_detector.py` — PostToolUse hook (`check-hook` stdin mode) + manual `check`, `sync`, `scan` subcommands
+  - `bootstrap.py` — shared registry CRUD and project root resolution via git
+- **Service registry**: `.claude/skills/service-registry.json` with territory globs, skill path, last sync
+- **Git hooks** (`pre-commit`, `pre-push`): idempotent marker-based installation for SSOT reminder and skill staleness warning
+- **Installer** (`install-service-skills.py`): single-purpose ~90-line script; copies trinity, merges settings.json hooks, activates git hooks; idempotent
+- **Phase 3 — Hook Registration**: new phase in `creating-service-skills` workflow verifies PreToolUse wiring, confirms territory globs in registry, communicates auto-activation to user
+
+### Changed
+- Project structure: moved into `project-skills/service-skills-set/` with `.claude/` subdirectory
+- `settings.json` PostToolUse hook moved to project-level (was only in skill frontmatter — now always-on)
+- PreToolUse added to `settings.json` for territory-based skill auto-enforcement
+
+### Fixed
+- `allowed-tools` in skill frontmatter: corrected to Claude Code native tool names — removed invalid MCP/Serena names
+- `SessionStart` removed from skill frontmatter (unsupported); moved to `settings.json`
+- Removed `disable-model-invocation: true` from workflow skill and scaffolder template
+- `project_root.glob()` type error in `bootstrap.py` fixed by wrapping with `Path()`
+
+### Documentation
+- Added `project-skills/service-skills-set/service-skills-readme.md`
+- New SSOT memory: `ssot_jaggers-agent-tools_service_skills_set_2026-02-23`
+
+---
+
+## [1.4.0] - 2026-02-23
+
+### Changed
+
+#### Delegating Skill Hardening
+- **Description rewrite**: Proactive language with trigger keywords (`tests`, `typos`, `refactors`, `code reviews`, `debugging`) — auto-discovery now fires without explicit "delegate" keyword
+- **Frontmatter cleanup**: Removed unsupported fields (`version`, `gemini-command`, `gemini-prompt`); added `allowed-tools: Bash`
+- **CCS nested session fix**: All CCS execution commands now use `env -u CLAUDECODE ccs {profile} -p "{task}"` — confirmed working inside Claude Code sessions
+- **Interactive menu**: Replaced TypeScript `ask_user()` pseudocode with prose `AskUserQuestion` instructions
+
+#### skill-suggestion.py Hook
+- **Orchestration patterns**: Added `ORCHESTRATION_PATTERNS` — hook now fires for code reviews, feature implementation, debugging, security audits, commit validation
+- **CLAUDECODE detection**: Hints correctly say "Gemini or Qwen directly" when running inside Claude Code (CCS unavailable), "CCS backend" otherwise
+- **Security exclusion fix**: Narrowed `security` exclude pattern to only block auth/vuln *implementation* — security *reviews* now correctly route to orchestration
+
+### Files Modified
+- `skills/delegating/SKILL.md` — Description, frontmatter, pseudocode, CCS command
+- `hooks/skill-suggestion.py` — Orchestration patterns, CLAUDECODE detection, security exclusion
+
+### Documentation
+- Updated SSOT: `ssot_cli_hooks_2026-02-03` → v1.1.0
+- New SSOT: `ssot_jaggers-agent-tools_delegating_skill_2026-02-23` v1.0.0
+
+---
+
+## [1.3.0] - 2026-02-22
+
+### Added
+
+#### CLI UX Improvements (vsync-inspired)
+- **Ora Spinners**: Visual feedback for all async operations (detect, diff, sync)
+- **Enhanced Status**: Last sync time, item counts, health indicators, actionable hints
+- **Single Confirmation**: Collect all changesets, display full plan, ask once
+- **Drifted Items Feedback**: Report skipped drifted items post-sync with backport hint
+
+### Changed
+
+#### Safety Improvements
+- **Prune Mode Guard**: Added `PruneModeReadError` — aborts if system read fails in prune mode
+- **Repo Root Detection**: Dynamic detection via `findRepoRoot()` utility (walks up looking for `skills/` + `hooks/`)
+- **Dry-Run Banner**: Moved from before target selection to after plan display
+- **Error Handling**: Global handlers for clean error messages (no stack traces)
+- **Ignored Items**: Filter `__pycache__`, `.DS_Store`, `node_modules` from diff scanning
+
+### Dependencies
+- Added `ora` for spinner UI
+
+### Files Modified
+- `cli/src/core/diff.ts` — Prune guard, ignored items filtering
+- `cli/src/utils/repo-root.ts` — New utility
+- `cli/src/commands/sync.ts` — Spinners, single confirm, feedback improvements
+- `cli/src/commands/status.ts` — Enhanced output with timestamps
+- `cli/src/core/manifest.ts` — Added `getManifestPath()`
+- `cli/src/index.ts` — Global error handlers
+
+### Documentation
+- New SSOT: `ssot_cli_ux_improvements_2026-02-22.md`
+
+---
+
+## [1.2.0] - 2026-02-21
+
+### Added
+
+#### CLI: TypeScript Migration
+- **Full TypeScript rewrite** of `cli/` — all modules ported from plain JavaScript ESM to strict TypeScript
+- **Commander.js** replaces `minimist` for structured sub-command routing
+- **Zod schemas** for runtime validation of `ChangeSet`, `SyncMode`, `Manifest`, `MCPServer`
+- **Adapter Pattern** — `ToolAdapter` base class with `ClaudeAdapter`, `GeminiAdapter`, `QwenAdapter` implementations
+  - `detectAdapter(systemRoot)` factory replaces scattered `includes('.claude')` checks codebase-wide
+- **Rollback protection** — `core/rollback.ts` backs up every file before write; restores all on any failure
+- **Hash-only diffing** — Pure MD5 comparison via `utils/hash.ts`; mtime used only as drift tie-breaker
+- **`prepare` npm script** — auto-builds on `npm install`, restoring `npx github:Jaggerxtrm/jaggers-agent-tools` support
+- **`vitest` test infrastructure** added to devDependencies (tests deferred, see `docs/plans/cli-testing.md`)
+
+#### New sub-commands
+- `jaggers-config sync [--dry-run] [-y] [--prune] [--backport]` — main sync
+- `jaggers-config status` — read-only diff view (no file writes)
+- `jaggers-config reset` — replaces `--reset` flag from old CLI
+
+#### Windows Compatibility (baked in)
+- `registry.ts` normalises backslashes before path matching
+- `config-adapter.ts` uses `python` (not `python3`) on Windows for hook scripts
+- `sync-executor.ts` falls back from symlinks to copy on Windows with a user warning
+
+### Changed
+- `cli/package.json` `bin` and root `package.json` `bin` now point to `cli/dist/index.js` (compiled output)
+- `cli/package.json` `scripts` updated: `build` (tsup), `dev` (tsx), `typecheck` (tsc), `test` (vitest), `start` (node dist)
+- Old `cli/index.js` and `cli/lib/*.js` preserved on disk but no longer referenced
+
+### Fixed
+- **Double-shebang bug** in tsup output — removed `banner` config, relying on tsup's auto-detection from `src/index.ts`
+
+---
+
+## [1.1.1] - 2026-02-03
+
+### Added
+- **Orchestrating Agents Skill**: Multi-model collaboration skill for Gemini and Qwen.
+- **Handshaking Workflows**: Deep multi-turn loops (Collaborative Design, Adversarial Review, Troubleshoot Session).
+- **Gemini Command Sync**: CLI support for synchronizing `.toml` commands and auto-generating them from skills.
+- **Cross-Agent Interactivity**: Support for both Gemini (`ask_user`) and Claude (`AskUserQuestion`) interactive menus.
+- Implement specialized Gemini slash commands (/delegate, /document, /prompt)
+- Enable zero-cloning installation via npx github:Jaggerxtrm/jaggers-agent-tools
+- Implement Vault Sync Architecture for non-destructive settings management. Protects local secrets, MCP servers, and auth data during sync. Includes atomic writes and dry-run mode.
+- **Architecture Roadmap**: Document CLI architectural improvements in ROADMAP.md based on multi-agent orchestration findings (Transactional Sync, Manifest Versioning, Namespace Prefixes, Observability).
+
+### Changed
+- **CLI Enhancement**: Automatically transforms `SKILL.md` into Gemini `.toml` command files during sync.
+- **Hook Migration**: Refined hook transformation logic for cross-agent compatibility.
+- Update SSOT and CHANGELOG for cross-agent compatibility and CLI improvements
+- Consolidate all v1.1.0 improvements: Zero-Cloning, Metadata-driven commands, and multi-turn orchestration
+- **ROADMAP.md**: Added "CLI Architecture Improvements" section with 5 phases addressing transactional sync, versioning, collision detection, observability, and transformation refactoring.
+
+### Fixed
+- Fix hook execution timeouts by updating settings.json to use milliseconds and enhancing transform-gemini.js to handle unit mismatches and improve hook naming.
+- Prevent redundant auto-generation of commands for core skills in CLI
+- Fix hardcoded paths in settings.json during sync
+- Fix ReferenceError in sync.js by adding missing import and verify via Qwen handshake
+
+---
+
+## [6.0.0] - 2026-02-01
+
+### Added
+
+#### `delegating` Skill (Unified)
+- **New `delegating` skill** replaces `ccs-delegation`
+- **Unified Backends**: Supports both CCS (cost-optimized) and unitAI (multi-agent workflows)
+- **Configuration-Driven**: All logic defined in `config.yaml`
+- **Auto-Focus**: Detects security/performance/quality focus from keywords
+- **Autonomous Workflow Selection**: Claude picks optimal unitAI workflow based on patterns
+
+### Removed
+
+#### `ccs-delegation` Skill
+- **Deprecated**: Fully replaced by `delegating` skill
+- **Removed**: `skills/ccs-delegation` directory deleted
+
+### Changed
+
+#### Skill Suggestions Hook
+- **Updated**: Suggests `/delegation` instead of `/ccs-delegation`
+- **Renamed**: `skill-suggestion.sh` → `skill-suggestion.py` for Python implementation
+
+---
+
+## [5.1.0] - 2026-01-30
+
+### Changed
+
+#### Naming Convention Alignment
+- **Skill `p` renamed to `prompt-improving`**
+  - Updated skill directory: `~/.claude/skills/p` → `~/.claude/skills/prompt-improving`
+  - Updated YAML frontmatter: `name: p` → `name: prompt-improving`
+  - Updated trigger syntax: `/p` → `/prompt-improving`
+  - Updated hook suggestions to reference `/prompt-improving`
+  - Follows Claude's naming convention with `-ing` suffix for improved clarity
+
+#### Breaking Changes
+- **`/p` command no longer works** - Use `/prompt-improving` instead
+- Users with muscle memory for `/p` will need to adapt to `/prompt-improving`
+- Hook suggestions now display `/prompt-improving` in systemMessage
+
+#### Migration Guide (5.0.0 → 5.1.0)
+**For Users:**
+- Replace all `/p "prompt"` invocations with `/prompt-improving "prompt"`
+- Update any documentation or workflows referencing the `/p` skill
+
+**For Backward Compatibility (Optional):**
+If you prefer to keep `/p` working via symlink:
+```bash
+ln -s ~/.claude/skills/prompt-improving ~/.claude/skills/p
+```
+
+---
+
+## [5.0.0] - 2026-01-30
+
+### Added
+
+#### Skills Enhancement
+- **UserPromptSubmit Hook** (`~/.claude/hooks/skill-suggestion.sh`)
+  - Proactive skill suggestions for `/p` and `/ccs` based on prompt analysis
+  - Bilingual pattern matching (Italian + English)
+  - Flexible synonym detection (e.g., "correggi|fix|sistema|repair")
+  - Sub-100ms execution time, no LLM calls
+  - Opt-in configuration via `settings.json`
+  - Detects simple tasks (typo, test, refactor, docs) → suggests `/ccs`
+  - Detects short/generic prompts → suggests `/p` for structure
+
+#### Configuration
+- **skillSuggestions config** in `settings.json`
+  - `enabled: true` - Hook active by default
+  - Can be disabled without restart
+- **UserPromptSubmit hook registration** in `settings.json`
+  - Timeout: 1s
+  - Command: `/home/dawid/.claude/hooks/skill-suggestion.sh`
+
+#### Skill Features
+- **AskUserQuestion dialogs** in `ccs-delegation` skill for interactive delegation choice
+- **AskUserQuestion clarification** in `p` skill for ambiguous prompts (<8 words)
+
+### Changed
+
+#### Skill `p` (Prompt Improver)
+- **SKILL.md**: Reduced from 118 to 64 lines (-46% size)
+- **Simplified context detection**: From 10 categories to 3 (ANALYSIS, DEV, REFACTOR)
+- **Removed multi-iteration improvement loop**: Single-pass processing only
+- **Inline scoring heuristics**: Replaced complex quality metrics with simple keyword checks
+- **Reference structure**: Merged prefill patterns into `xml_core.md` (+20 lines)
+
+#### Skill `ccs-delegation`
+- **SKILL.md**: Reduced from 486 to 151 lines (-69% size)
+- **Keyword-based profile selection**: Replaced quantitative complexity scoring (0-10 scale)
+  - Simple patterns: `typo|test|doc` → glm
+  - Reasoning patterns: `analiz|think|reason` → gemini
+  - Architecture patterns: `architecture|entire|codebase` → gemini
+- **Bilingual support**: IT+EN keywords throughout (e.g., "correggi|fix", "aggiungi.*test|add.*test")
+- **Simplified execution flow**: Detect → Ask → Select Profile → Execute (removed fallback chains)
+
+#### Performance Improvements
+- **Skill load time**: 5-8s → <1s (-80-85% reduction)
+- **Total token overhead**: 155KB → 16KB (-90% reduction)
+- **Pattern matching**: Extended from basic English to IT+EN with wildcards
+
+### Removed
+
+#### Skill `p` References (46KB total)
+- `quality_metrics.md` (12.7KB, 511 lines) - Complex 0-100 scoring system
+- `context_detection_rules.md` (10.4KB) - 10-category detection rules
+- `prefill_patterns.md` (10KB) - Standalone prefill examples (merged into xml_core.md)
+- `before_after_examples.md` (12.9KB) - Redundant examples
+
+#### Skill `ccs-delegation` References (95KB total)
+- `task_complexity_scoring.md` (14.4KB, 478 lines) - Quantitative complexity algorithm
+- `smart_context_gathering.md` (16.6KB, 643 lines) - Multi-level context system
+- `fallback_chain.md` (15.5KB) - Edge-case fallback handling
+- `parallel_delegation.md` (17.1KB) - Multi-agent parallel execution
+- `delegation_history_analysis.md` (15.7KB) - Learning/persistence system
+
+### Fixed
+
+#### Pattern Matching
+- **Too rigid English-only patterns** → Extended to bilingual IT+EN with synonyms
+- **Missing common terms** → Added: "rimuovi|remove", "modifica|modify", "sistema|repair"
+- **Case sensitivity issues** → All patterns use case-insensitive matching (`grep -i`)
+
+#### Hook Configuration
+- **Hook script not executable** → Added `chmod +x` to deployment checklist
+- **Missing skillSuggestions config** → Added to `settings.json` with `enabled: true`
+
+---
+
+## [4.2.0] - Pre-refactoring baseline
+
+### Changed
+#### Skills State Before Refactoring
+- **Skill `p`**: 118 lines, 52KB references (9 files)
+- **Skill `ccs-delegation`**: 486 lines, 103KB references (6 files)
+- **Total overhead**: 155KB token cost per skill activation
+- **Load time**: 5-8 seconds per skill invocation