oh-my-customcode 0.59.1 → 0.60.1

package/README.md

@@ -13,7 +13,7 @@
 
 **[한국어 문서 (Korean)](./README_ko.md)**
 
-46 agents. 95 skills. 21 rules. One command.
+46 agents. 97 skills. 21 rules. One command.
 
 ```bash
 npm install -g oh-my-customcode && cd your-project && omcustom init
@@ -146,7 +146,7 @@ Each agent declares its tools, model, memory scope, and limitations in YAML fron
 
 ---
 
-### Skills (95)
+### Skills (97)
 
 | Category | Count | Includes |
 |----------|-------|----------|
@@ -282,7 +282,7 @@ your-project/
 ├── CLAUDE.md                   # Entry point
 ├── .claude/
 │   ├── agents/                 # 46 agent definitions
-│   ├── skills/                 # 95 skill modules
+│   ├── skills/                 # 97 skill modules
 │   ├── rules/                  # 21 governance rules (R000-R021)
 │   ├── hooks/                  # 15 lifecycle hook scripts
 │   ├── schemas/                # Tool input validation schemas

package/dist/cli/index.js

@@ -9325,7 +9325,7 @@ var init_package = __esm(() => {
     workspaces: [
       "packages/*"
     ],
-    version: "0.59.1",
+    version: "0.60.1",
     description: "Batteries-included agent harness for Claude Code",
     type: "module",
     bin: {

package/dist/index.js

@@ -1672,7 +1672,7 @@ var package_default = {
   workspaces: [
     "packages/*"
   ],
-  version: "0.59.1",
+  version: "0.60.1",
   description: "Batteries-included agent harness for Claude Code",
   type: "module",
   bin: {

package/package.json

@@ -3,7 +3,7 @@
   "workspaces": [
     "packages/*"
   ],
-  "version": "0.59.1",
+  "version": "0.60.1",
   "description": "Batteries-included agent harness for Claude Code",
   "type": "module",
   "bin": {

package/templates/.claude/rules/MUST-agent-design.md

@@ -32,6 +32,7 @@ escalation:              # Model escalation policy (optional)
   threshold: 2           # Failures before advisory
 soul: true                 # Enable SOUL.md identity injection
 isolation: worktree | sandbox  # worktree = git worktree, sandbox = restricted bash
+sandboxFailIfUnavailable: true  # Exit if sandbox unavailable (v2.1.83+)
 background: true           # Run in background
 maxTurns: 10               # Max conversation turns
 maxTokens: 100000          # Per-turn token ceiling
@@ -39,6 +40,7 @@ mcpServers: [server-1]     # MCP servers available
 hooks:                     # Agent-specific hooks
   PreToolUse:
     - matcher: "Edit"
+      if: "Edit(*.md)"      # Conditional filter (permission rule syntax, v2.1.85+)
       command: "echo hook"
 permissionMode: bypassPermissions  # Permission mode
 disallowedTools: [Bash]    # Tools to disallow
@@ -48,7 +50,7 @@ limitations:               # Negative capability declarations
 domain: backend              # backend | frontend | data-engineering | devops | universal
 ```
 
-> **Note**: `isolation`, `background`, `maxTurns`, `maxTokens`, `mcpServers`, `hooks`, `permissionMode`, `disallowedTools`, `limitations` are supported in Claude Code v2.1.63+. Hook types `PostCompact`, `Elicitation`, `ElicitationResult` require v2.1.76+.
+> **Note**: `isolation`, `background`, `maxTurns`, `maxTokens`, `mcpServers`, `hooks`, `permissionMode`, `disallowedTools`, `limitations` are supported in Claude Code v2.1.63+. Hook types `PostCompact`, `Elicitation`, `ElicitationResult` require v2.1.76+. `CwdChanged`, `FileChanged` hook events and `managed-settings.d/` drop-in directory require v2.1.83+. Conditional `if` field for hooks requires v2.1.85+.
 
 <!-- DETAIL: Isolation/Token/Limitations/Escalation details
 ### Isolation Modes

package/templates/.claude/rules/MUST-agent-teams.md

@@ -25,6 +25,15 @@ Available when `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` or TeamCreate/SendMessag
 
 **When Agent Teams is enabled and criteria are met, usage is required.**
 
+### Scope: Intra-Session vs Cross-Session
+
+| Scope | Tool | Protocol | Use Case |
+|-------|------|----------|----------|
+| Intra-session | `SendMessage` (Agent Teams) | Peer-to-peer within team | Multi-agent collaboration in one session |
+| Cross-session | `send_message` (claude-peers-mcp) | Broker-mediated | Multi-terminal/project coordination |
+
+These are distinct mechanisms. Agent Teams `SendMessage` requires `TeamCreate` and operates within a single Claude Code session. claude-peers-mcp `send_message` operates across separate Claude Code processes via a localhost broker.
+
 ## Self-Check (Before Agent Tool)
 
 Before using Agent tool for 2+ agent tasks, complete this check:

package/templates/.claude/skills/action-validator/SKILL.md

@@ -0,0 +1,75 @@
+---
+name: action-validator
+description: Pre-action boundary checking — validates agent tool calls against declared capabilities and task contracts
+scope: core
+user-invocable: false
+---
+
+# Action Validator Skill
+
+## Purpose
+
+Advisory pre-action validation layer that checks agent tool calls against declared capabilities, file access scope (R002), and task contracts before execution. Inspired by AutoHarness (Google DeepMind) — enforcing action-space legality at agent boundaries.
+
+This skill does NOT block actions (R021 advisory-first model). It emits warnings when agents attempt operations outside their declared scope.
+
+## Validation Checks
+
+| Check | What | Against |
+|-------|------|---------|
+| Tool scope | Tool being called | Agent's `tools` frontmatter list |
+| File scope | File path in Write/Edit | R002 file access rules |
+| Domain scope | Target file extension | Agent's `domain` frontmatter |
+| Task contract | Operation type | Task description constraints |
+
+## Advisory Format
+
+```
+--- [Action Validator] Scope warning ---
+  Agent: {agent-name}
+  Tool: {tool-name}
+  Target: {file-path}
+  Issue: {description}
+  Declared scope: {agent's declared tools/domain}
+  💡 Suggestion: {recommended action}
+---
+```
+
+## Integration Points
+
+| System | How |
+|--------|-----|
+| PreToolUse hooks | Optional hook to check tool calls (advisory only) |
+| pipeline-guards | Complements pipeline stage gates |
+| adversarial-review | Provides action-space-legality criterion |
+| R002 (Permissions) | Validates against declared file access rules |
+| R010 (Orchestrator) | Orchestrator validates subagent scope claims |
+
+## Policy Cache Pattern
+
+For high-repetition agents (e.g., mgr-gitnerd commit workflows), capture validated decision paths as reusable policies:
+
+```yaml
+policy_cache:
+  agent: mgr-gitnerd
+  action: git-commit
+  validated_steps:
+    - tool: Bash
+      pattern: "git add *"
+      verdict: allow
+    - tool: Bash
+      pattern: "git commit *"
+      verdict: allow
+    - tool: Bash
+      pattern: "git push *"
+      verdict: warn_confirm
+```
+
+Policy caching reduces redundant LLM calls for well-understood workflows. Policies are advisory — the orchestrator may override.
+
+## Scope
+
+This skill is an advisory layer, not a hard enforcement mechanism:
+- **Does**: Emit warnings, log scope violations, suggest corrections
+- **Does NOT**: Block tool execution, modify agent behavior, override R021
+- **Future**: May integrate with PreToolUse hooks for automated checking (see R021 promotion criteria)

package/templates/.claude/skills/adversarial-review/SKILL.md

@@ -70,3 +70,11 @@ Fix: Recommended remediation
 - Complements `dev-review` (best practices) with attacker perspective
 - Works with `sec-codeql-expert` for pattern-based + logic-based coverage
 - Can be chained: `dev-review` → `adversarial-review` for complete coverage
+- Works with `action-validator` for action-space legality checking
+
+### Action-Space Legality (AutoHarness Pattern)
+
+- [ ] Do agents only call tools within their declared `tools` frontmatter?
+- [ ] Do file operations stay within R002-declared access scope?
+- [ ] Are domain boundaries respected (backend agent not editing frontend files)?
+- [ ] Could an agent's task contract be tightened without losing functionality?

package/templates/.claude/skills/evaluator-optimizer/SKILL.md

@@ -33,6 +33,27 @@ evaluator-optimizer:
   max_iterations: 3               # Default, hard cap: 5
 ```
 
+### Pre-Negotiation (Sprint Contract Pattern)
+
+Optional phase where generator and evaluator agree on rubric interpretation before the first iteration. Inspired by Anthropic's harness design for long-running applications.
+
+```yaml
+evaluator-optimizer:
+  pre_negotiation:
+    enabled: true              # Default: false
+    rounds: 1                  # Negotiation rounds (1-2)
+  generator:
+    agent: fe-design-expert
+    ...
+```
+
+When enabled:
+1. Generator receives the rubric and proposes its interpretation + planned approach
+2. Evaluator reviews and may adjust rubric weights or add clarifications
+3. Both proceed with aligned expectations, reducing wasted iterations
+
+Use when: tasks requiring 3+ iterations consistently, or when generator-evaluator score disagreements exceed 0.3.
+
 ### Parameter Details
 
 | Parameter | Required | Default | Description |
@@ -307,3 +328,38 @@ When ecomode is active (R013), compress output:
 - The evaluator prompt MUST include the full rubric to ensure consistent scoring
 - Iteration state (best score, best output) is tracked by the orchestrator
 - The hard cap of 5 iterations prevents runaway refinement loops
+
+## Domain Examples
+
+### UI Generation (Anti-AI-Slop)
+
+For UI/design generation tasks, use weighted rubrics that penalize generic AI patterns:
+
+```yaml
+evaluator-optimizer:
+  generator:
+    agent: fe-design-expert
+    model: sonnet
+  evaluator:
+    agent: fe-design-expert
+    model: opus
+  rubric:
+    - criterion: originality
+      weight: 0.40
+      description: "No stock patterns (centered hero + 3-card grid). Unique layout, typography choices, color relationships."
+    - criterion: craft
+      weight: 0.35
+      description: "Intentional spacing, consistent type scale, purposeful color usage. Details that show care."
+    - criterion: functionality
+      weight: 0.25
+      description: "Accessibility (WCAG 2.1 AA), responsive behavior, interaction states."
+  quality_gate:
+    type: score_threshold
+    threshold: 0.85
+  pre_negotiation:
+    enabled: true
+```
+
+Weight ordering (originality > craft > functionality) follows Anthropic's anti-slop principle: functionality is table stakes, but originality and craft distinguish quality output from generic AI generation.
+
+Integration: Works with [impeccable-design](/skills/impeccable-design) skill for design language enforcement.

package/templates/.claude/skills/monitoring-setup/SKILL.md

@@ -114,3 +114,32 @@ OTEL_LOGS_EXPORTER=otlp
 OTEL_EXPORTER_OTLP_PROTOCOL=grpc
 OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317
 ```
+
+## HTTP-Level Inspection (Optional)
+
+For deeper payload-level debugging beyond aggregated metrics, [Claude Inspector](https://github.com/kangraemin/claude-inspector) provides MITM proxy inspection of Claude Code HTTP traffic.
+
+| Aspect | OTel Monitoring (this skill) | Claude Inspector |
+|--------|------------------------------|-----------------|
+| Layer | Application (hooks, stdout) | HTTP (MITM proxy) |
+| Metrics | Aggregated (cost, tokens, duration) | Per-request payload breakdown |
+| Cache visibility | Not available | Prompt Cache hit/miss rates |
+| Sub-agent view | Summary via hooks | Full parent vs sub-agent context comparison |
+| Setup | Built-in (hooks + statusline) | External tool (Homebrew on macOS) |
+
+### When to Use
+
+- **OTel monitoring**: Daily operations, cost tracking, performance trends
+- **Claude Inspector**: Debugging specific payload issues, measuring CLAUDE.md token impact, verifying ecomode (R013) effectiveness, profiling sub-agent context inheritance
+
+### Setup
+
+```bash
+# macOS
+brew install kangraemin/tap/claude-inspector
+
+# Run proxy
+claude-inspector
+```
+
+Claude Inspector is external to oh-my-customcode and does not require any project configuration changes.

package/templates/.claude/skills/peer-messaging/SKILL.md

@@ -0,0 +1,59 @@
+---
+name: peer-messaging
+description: Cross-session Claude Code instance messaging via claude-peers-mcp broker
+scope: core
+user-invocable: false
+---
+
+# Peer Messaging Skill
+
+## Purpose
+
+Enables cross-session coordination between multiple Claude Code instances through the claude-peers-mcp broker. Complements Agent Teams (R018, intra-session) with inter-session messaging.
+
+## Scope Clarification
+
+| Scope | Mechanism | Tools | Use Case |
+|-------|-----------|-------|----------|
+| Intra-session agents | Agent Teams (R018) | TeamCreate, SendMessage | Single session multi-agent collaboration |
+| Cross-session instances | claude-peers-mcp | list_peers, send_message | Multi-terminal/project real-time coordination |
+| Cross-session memory | claude-mem | save_memory, search | Async memory persistence |
+
+> **Important**: R018's `SendMessage` and claude-peers-mcp's `send_message` are different tools with different scopes. Do not confuse them.
+
+## MCP Tool Mapping
+
+| Tool | Purpose | oh-my-customcode Scenario |
+|------|---------|---------------------------|
+| `list_peers` | Discover active Claude instances | `omcustom:status` system overview |
+| `send_message` | Send message to peer | Cross-project workflow coordination |
+| `set_summary` | Broadcast current task summary | DAG cross-project step sync |
+| `check_messages` | Read incoming messages | Receive coordination signals |
+
+## Use Cases
+
+### Multi-Project Workflow
+Terminal A runs `auto-dev` on project-1; Terminal B works on dependent project-2. Peers coordinate via messages when blocking dependencies are resolved.
+
+### Cross-Project QA
+Share test infrastructure state between projects running concurrent test suites.
+
+### DAG Bridge
+`dag-orchestration` cross-project steps can use peer messaging for synchronization (currently impossible without this tool).
+
+## Setup
+
+```bash
+# Install broker (optional MCP server)
+npm install -g claude-peers-mcp
+
+# Add to MCP config
+claude mcp add claude-peers-mcp -- npx claude-peers-mcp
+```
+
+## Integration
+
+- Works with R018 Agent Teams (different scope, complementary)
+- Works with claude-mem (async vs sync messaging)
+- Works with `omcustom:status` (peer discovery)
+- Broker runs on localhost:7899 (SQLite-backed)

package/templates/.claude/skills/stuck-recovery/SKILL.md

@@ -25,6 +25,7 @@ Detects when tasks are stuck in repetitive failure loops and advises recovery st
 | Model escalation | Agent retry loop | Trigger model-escalation advisory |
 | Alternative approach | Edit loop detected | Suggest different file/method |
 | Human intervention | All automated strategies exhausted | Ask user for guidance |
+| Context reset | Long-running task (>30min) or context >80% | Structured handoff: save state to memory, create fresh session with task summary |
 
 ## Architecture
 
@@ -53,3 +54,26 @@ PostToolUse (Edit, Write, Bash, Task) → stuck-detector.sh
 - Respects R010 (advisory only, orchestrator decides)
 - Uses same PPID-scoped temp file pattern as other hooks
 - Works with task-outcome-recorder.sh data when available
+
+## Context Reset Strategy
+
+For long-running tasks (>30 minutes) or when context usage exceeds 80%, context reset is preferred over compaction:
+
+1. **Save state**: Write current progress, decisions, and open items to native auto-memory
+2. **Create handoff**: Generate structured task summary with:
+   - Completed steps and their outcomes
+   - Current step and its state
+   - Remaining steps
+   - Key decisions made and their rationale
+3. **Reset**: Start fresh session with handoff document as input
+
+Context reset preserves decision quality by avoiding the information loss inherent in compaction. Based on Anthropic's finding that models experience "context anxiety" — prematurely concluding tasks due to perceived token limits.
+
+### When to Use
+
+| Condition | Strategy |
+|-----------|----------|
+| Context < 60% | Continue normally |
+| Context 60-80% | Consider `/compact` |
+| Context > 80% OR duration > 30min | Context reset recommended |
+| Repeated compaction in same session | Context reset required |

package/templates/.claude/skills/workflow/SKILL.md

@@ -1,20 +1,20 @@
 ---
-name: workflow
-description: Invoke YAML-defined workflows by name — /workflow omcustom-dev runs the full pipeline
+name: omcustom:workflow
+description: Invoke YAML-defined workflows by name — /omcustom:workflow auto-dev runs the full pipeline
 scope: harness
 user-invocable: true
 effort: high
 argument-hint: "<workflow-name> | (no args to list available)"
 ---
 
-# /workflow — Workflow Invocation
+# /omcustom:workflow — Workflow Invocation
 
 ## Usage
 
 ```
-/workflow omcustom-dev     # Run the omcustom-dev workflow
-/workflow                  # List available workflows
-/workflow:resume           # Resume a halted workflow
+/omcustom:workflow auto-dev     # Run the auto-dev workflow
+/omcustom:workflow                  # List available workflows
+/omcustom:workflow:resume           # Resume a halted workflow
 ```
 
 ## Behavior
@@ -24,7 +24,7 @@ argument-hint: "<workflow-name> | (no args to list available)"
 Scan `workflows/*.yaml` and display:
 ```
 Available workflows:
-  omcustom-dev — verify-done issues release batch: triage → plan → implement → verify → PR
+  auto-dev — verify-done issues release batch: triage → plan → implement → verify → PR
 ```
 
 ### Run Mode (with workflow name)
@@ -35,7 +35,7 @@ Available workflows:
 4. Invoke workflow-runner skill with the loaded definition
 5. Report completion or failure
 
-### Resume Mode (/workflow:resume)
+### Resume Mode (/omcustom:workflow:resume)
 
 1. Check for state file: `/tmp/.claude-workflow-*-{PPID}.json`
 2. If found: show halted workflow name and failed step

package/templates/CLAUDE.md

@@ -138,7 +138,7 @@ project/
 +-- CLAUDE.md                    # 진입점
 +-- .claude/
 |   +-- agents/                  # 서브에이전트 정의 (46 파일)
-|   +-- skills/                  # 스킬 (95 디렉토리)
+|   +-- skills/                  # 스킬 (97 디렉토리)
 |   +-- rules/                   # 전역 규칙 (R000-R021)
 |   +-- hooks/                   # 훅 스크립트 (보안, 검증, HUD)
 |   +-- contexts/                # 컨텍스트 파일 (ecomode)

package/templates/manifest.json

@@ -1,5 +1,5 @@
 {
-  "version": "0.59.1",
+  "version": "0.60.1",
   "lastUpdated": "2026-03-24T00:00:00.000Z",
   "components": [
     {
@@ -18,7 +18,7 @@
       "name": "skills",
       "path": ".claude/skills",
       "description": "Reusable skill modules (includes slash commands)",
-      "files": 95
+      "files": 97
     },
     {
       "name": "guides",

package/templates/workflows/{omcustom-dev.yaml → auto-dev.yaml}

@@ -1,8 +1,8 @@
-# /workflow:omcustom-dev — Full-auto release pipeline
-# Collects verify-done issues → triage → plan → implement → verify → PR
+# /omcustom:workflow auto-dev — Full-auto release pipeline
+# Collects verify-done issues → triage → plan → implement → verify → PR → followup
 
-name: omcustom-dev
-description: "verify-done issues release batch: triage → plan → implement → verify → PR"
+name: auto-dev
+description: "verify-done issues release batch: triage → plan → implement → verify → PR → followup"
 mode: auto
 error: halt-and-report
 
@@ -33,3 +33,7 @@ steps:
   - name: release
     action: create-pr
     description: Create release branch and pull request
+
+  - name: followup
+    skill: post-release-followup
+    description: Recommend follow-up actions — user chooses to execute now or register as issues