@slamb2k/mad-skills 2.0.47 → 2.0.49

package/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "mad-skills",
   "description": "AI-assisted planning, development and governance tools",
-  "version": "2.0.47",
+  "version": "2.0.49",
   "author": {
     "name": "slamb2k",
     "url": "https://github.com/slamb2k"

package/hooks/lib/config.cjs

@@ -62,8 +62,5 @@ module.exports = {
         'Read', 'Glob', 'Grep', 'ToolSearch', 'Agent', 'WebSearch', 'WebFetch',
       ],
     },
-    hookify: {
-      rulePattern: /^hookify\..*\.local\.md$/,
-    },
   },
 };

package/hooks/lib/plugin-health.cjs

@@ -2,7 +2,6 @@
 
 const { join } = require('path');
 const { homedir } = require('os');
-const { readdirSync } = require('fs');
 const config = require('./config.cjs');
 const state = require('./state.cjs');
 const { readJson } = require('./utils.cjs');
@@ -11,7 +10,6 @@ const { readJson } = require('./utils.cjs');
  * Plugin Health — detect performance anti-patterns in companion plugins.
  *
  * Checks:
- *   - Hookify enabled with no rules (pure overhead)
  *   - claude-mem missing SKIP_TOOLS for read-only tools
  *   - claude-mem context injection too high (when OMC also active)
  *
@@ -27,38 +25,9 @@ function checkPluginHealth(projectDir, output) {
 
   const plugins = settings.enabledPlugins || {};
 
-  checkHookify(projectDir, plugins, output);
   checkClaudeMem(plugins, output);
 }
 
-// ─── hookify ──────────────────────────────────────────────────────────
-
-function checkHookify(projectDir, plugins, output) {
-  const hookifyKey = Object.keys(plugins).find(k => k.includes('hookify'));
-  if (!hookifyKey || plugins[hookifyKey] !== true) return; // Not enabled
-
-  // Count rule files in project .claude/ directory
-  let ruleCount = 0;
-  try {
-    const projectClaudeDir = join(projectDir, '.claude');
-    const files = readdirSync(projectClaudeDir);
-    ruleCount = files.filter(f => config.pluginHealth.hookify.rulePattern.test(f)).length;
-  } catch { /* directory doesn't exist — zero rules */ }
-
-  // Also check global .claude/ directory
-  try {
-    const globalClaudeDir = join(homedir(), '.claude');
-    const files = readdirSync(globalClaudeDir);
-    ruleCount += files.filter(f => config.pluginHealth.hookify.rulePattern.test(f)).length;
-  } catch { /* noop */ }
-
-  if (ruleCount === 0) {
-    output.signals.push(
-      '\u26A0 Hookify enabled but no rules configured \u2014 fires on every tool call for nothing. Run /brace or disable in settings',
-    );
-  }
-}
-
 // ─── claude-mem ───────────────────────────────────────────────────────
 
 function checkClaudeMem(plugins, output) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@slamb2k/mad-skills",
-  "version": "2.0.47",
+  "version": "2.0.49",
   "description": "Claude Code skills collection — full lifecycle development tools",
   "type": "module",
   "repository": {

package/skills/brace/SKILL.md

@@ -102,7 +102,7 @@ For each row, in order:
 **Plugin detection:** For plugin dependencies (Type = plugin), check
 `~/.claude/settings.json` → `enabledPlugins` for a key containing the plugin
 name set to `true`. Store results as `PLUGIN_STATE` (`claude_mem_installed`,
-`omc_installed`, `hookify_installed`) for use in Phase 4 and Phase 7.
+`omc_installed`) for use in Phase 4 and Phase 7.
 
 1. Capture **FLAGS** from the user's request
 
@@ -170,16 +170,15 @@ Store result as `upgrade_legacy: true|false` in USER_CONFIG.
 
 4. Ask installation level via AskUserQuestion:
 
-   Question: "Where should global preferences and universal principles go?"
+   Question: "Install universal guidance (preferences, principles) at user level?"
    Options:
-   - "Both — global + project (Recommended)" → portable AND global coverage
-   - "Global (~/.claude/CLAUDE.md) only" → applies to all projects
-   - "Project level only" → self-contained, portable
+   - "Yes, install globally (Recommended)" — applies to all projects via `~/.claude/CLAUDE.md`
+   - "No, project level only" — self-contained in this project's CLAUDE.md
 
 5. Store as USER_CONFIG:
    - project_name: from directory name or user override
    - description: from user input
-   - install_level: "both" | "global" | "project"
+   - install_level: "global" | "project"
 
 **If cancelled, stop here.**
 
@@ -220,7 +219,7 @@ Will upgrade: {list of items with status "upgrade"}
 Will remove:  {list of items with status "remove" — legacy cleanup}
 Will clean:   {list of items with status "cleanup" — remove legacy references}
 Will skip:    {count} existing items
-Global config: {install_level description}
+Universal guidance: {install_level description}
 
 Proceed?
 ```
@@ -247,7 +246,7 @@ Before sending the prompt, substitute these variables:
 - `{ACTION_PLAN}` — the action plan from Phase 3
 - `{PROJECT_NAME}` — from USER_CONFIG
 - `{PROJECT_DESCRIPTION}` — from USER_CONFIG
-- `{INSTALL_LEVEL}` — from USER_CONFIG ("both", "global", or "project")
+- `{INSTALL_LEVEL}` — from USER_CONFIG ("global" or "project")
 - `{CLAUDE_MD_TEMPLATE}` — read from `references/claude-md-template.md`
   (the section between BEGIN TEMPLATE and END TEMPLATE)
 - `{GITIGNORE_CONTENT}` — read from `assets/gitignore-template`
@@ -334,10 +333,10 @@ This phase modifies user-level settings files (`~/.claude/settings.json`,
 If `--force` is set, apply all recommendations without prompting.
 
 **Plugin presence guards:** Only audit plugins detected as installed during
-pre-flight (`PLUGIN_STATE`). Skip H1/H2 if hookify is absent. Skip M1/M2/M3
-if claude-mem is absent. Skip M2 if OMC is absent (M2 requires both). The
-`{PLUGIN_ROLE_SEPARATION}` content in CLAUDE.md is only injected when both
-claude-mem and OMC are confirmed enabled.
+pre-flight (`PLUGIN_STATE`). Skip M1/M2/M3 if claude-mem is absent. Skip M2
+if OMC is absent (M2 requires both). The `{PLUGIN_ROLE_SEPARATION}` content
+in CLAUDE.md is only injected when both claude-mem and OMC are confirmed
+enabled.
 
 If no companion plugins are installed at all, output:
 ```
@@ -383,8 +382,6 @@ If "Let me choose", present individual findings as multi-select.
 
 | Code | Check | Condition | Severity |
 |------|-------|-----------|----------|
-| H1 | Hookify: no rules | enabled + zero `hookify.*.local.md` files | high |
-| H2 | Hookify: python3 broken | enabled + `python3 --version` fails | high |
 | M1 | claude-mem: read-only tools | SKIP_TOOLS missing Read/Glob/Grep/ToolSearch/Agent/WebSearch/WebFetch | medium |
 | M2 | claude-mem: high context | observations > 10 or sessions > 3, AND OMC also enabled | medium |
 | M3 | claude-mem: provider=claude | provider is "claude" (SDK spawn known-broken) | low |

package/skills/brace/assets/global-preferences-template.md

@@ -1,8 +1,11 @@
 # Global Preferences & Universal Principles
 
 Template appended to `~/.claude/CLAUDE.md` by brace when the user
-selects "global" or "both" installation level. The Phase 4 agent inserts
-this content before the "## Current Skills" section.
+selects "global" install level. When the user selects "project" level,
+this file is not used — instead, the universal principles are written
+to the project CLAUDE.md (with a redundancy check against global).
+The Phase 4 agent inserts this content before the "## Current Skills"
+section.
 
 ---
 

package/skills/brace/references/claude-md-template.md

@@ -4,9 +4,10 @@ Template for the generated project CLAUDE.md. The Phase 4 agent substitutes
 `{VARIABLE}` placeholders and writes to the project root.
 
 `{UNIVERSAL_PRINCIPLES}` is populated with the Question & Assumption
-Accountability and Communication sections when install_level is "project"
-or "both". Left empty when install_level is "global" (principles are in
-`~/.claude/CLAUDE.md` instead).
+Accountability section when install_level is "project" AND those sections
+are not already present in `~/.claude/CLAUDE.md`. Left empty when
+install_level is "global" (principles are in the global config instead)
+or when the sections would be redundant with existing global content.
 
 ---
 

package/skills/brace/references/phase-prompts.md

@@ -146,21 +146,31 @@ skills-based workflow while preserving all other content.
   This project uses claude-mem for persistent cross-session memory.
   ```
 
-### Global preferences (conditional)
+### Global preferences and universal principles
 
-If INSTALL_LEVEL is "global" or "both":
+**If INSTALL_LEVEL is "global":**
 - Read ~/.claude/CLAUDE.md
 - If it does NOT contain "## Global Preferences", insert the Global
   Preferences Content (below) immediately before "## Current Skills"
 - If it already contains "## Global Preferences", skip (idempotent)
-
-### {UNIVERSAL_PRINCIPLES} substitution
-
-If INSTALL_LEVEL is "project" or "both", substitute {UNIVERSAL_PRINCIPLES}
-in the CLAUDE.md template with the Universal Principles Content below.
-
-If INSTALL_LEVEL is "global", substitute {UNIVERSAL_PRINCIPLES} with an
-empty string (principles are in the global config instead).
+- Substitute {UNIVERSAL_PRINCIPLES} in the project CLAUDE.md template with
+  an empty string (principles are in the global config instead).
+
+**If INSTALL_LEVEL is "project":**
+- Do NOT modify ~/.claude/CLAUDE.md.
+- Before writing universal principles into the project CLAUDE.md, check
+  ~/.claude/CLAUDE.md for existing equivalent sections (redundancy guard):
+  - If global contains "## Global Preferences" → SKIP that section in
+    project CLAUDE.md. Add to SCAFFOLD_REPORT.skipped_redundant.
+  - If global contains "## Universal Operating Principles" → SKIP that
+    section in project CLAUDE.md. Add to SCAFFOLD_REPORT.skipped_redundant.
+- For any sections NOT found in global, substitute {UNIVERSAL_PRINCIPLES}
+  in the project CLAUDE.md template with only the non-redundant sections
+  from the Universal Principles Content below.
+- If ALL sections are redundant, substitute {UNIVERSAL_PRINCIPLES} with
+  an empty string.
+- Report each skipped section to the user:
+  "⏭️ Skipped {section} in project CLAUDE.md — already present in ~/.claude/CLAUDE.md"
 
 ### CLAUDE.md Template
 
@@ -206,6 +216,7 @@ SCAFFOLD_REPORT:
   cleaned: [list of files cleaned of legacy references]
   preserved_content: [any key decisions extracted from memory/MEMORY.md, or empty]
   skipped: [list of items skipped]
+  skipped_redundant: [sections skipped in project CLAUDE.md because already in global, or empty]
   global_updated: true|false|skipped
   errors: [any errors encountered]
 ```
@@ -286,39 +297,19 @@ Limit PLUGIN_REPORT to 30 lines maximum.
    node -e "
      const s = JSON.parse(require('fs').readFileSync('$HOME/.claude/settings.json','utf8'));
      const p = s.enabledPlugins || {};
-     const hookify = Object.keys(p).find(k => k.includes('hookify'));
      const mem = Object.keys(p).find(k => k.includes('claude-mem'));
      const omc = Object.keys(p).find(k => k.includes('oh-my-claudecode'));
-     console.log('hookify_installed:' + !!hookify);
-     console.log('hookify_enabled:' + (hookify && p[hookify] === true));
      console.log('claude_mem_installed:' + !!mem);
      console.log('claude_mem_enabled:' + (mem && p[mem] === true));
      console.log('omc_installed:' + !!omc);
      console.log('omc_enabled:' + (omc && p[omc] === true));
    "
-   Record: hookify_installed/enabled, claude_mem_installed/enabled, omc_installed/enabled
+   Record: claude_mem_installed/enabled, omc_installed/enabled
 
    If a plugin is not installed (not in enabledPlugins at all), skip its
    entire audit section below and report all its fields as "N/A".
 
-2. **Hookify audit** (only if hookify_enabled == true)
-   a. Check python3 availability:
-      python3 --version >/dev/null 2>&1 && echo "python3_available:true" || echo "python3_available:false"
-      Record: python3_available
-
-   b. Count hookify rule files:
-      RULE_COUNT=0
-      for f in "$HOME/.claude"/hookify.*.local.md .claude/hookify.*.local.md; do
-        [ -f "$f" ] && RULE_COUNT=$((RULE_COUNT + 1))
-      done
-      echo "hookify_rule_count:$RULE_COUNT"
-      Record: hookify_rule_count
-
-   c. Determine findings:
-      - If python3_available == false → finding H2
-      - If hookify_rule_count == 0 → finding H1
-
-3. **claude-mem audit** (only if claude_mem_enabled == true)
+2. **claude-mem audit** (only if claude_mem_enabled == true)
    MEM_SETTINGS="$HOME/.claude-mem/settings.json"
    if [ -f "$MEM_SETTINGS" ]; then
      node -e "
@@ -349,10 +340,6 @@ Limit PLUGIN_REPORT to 30 lines maximum.
 
 PLUGIN_REPORT:
   settings_file_found: true|false
-  hookify_installed: true|false
-  hookify_enabled: true|false
-  hookify_python3_available: true|false|N/A
-  hookify_rule_count: {number}|N/A
   claude_mem_installed: true|false
   claude_mem_enabled: true|false
   claude_mem_skip_tools: {current value}|N/A
@@ -363,5 +350,5 @@ PLUGIN_REPORT:
   claude_mem_has_openrouter_key: true|false|N/A
   omc_installed: true|false
   omc_enabled: true|false
-  findings: {comma-separated list of H1,H2,M1,M2,M3 or "none"}
+  findings: {comma-separated list of M1,M2,M3 or "none"}
 ```

package/skills/brace/references/plugin-tuning-steps.md

@@ -6,28 +6,6 @@ checks current value before modifying.
 
 ---
 
-## H1/H2: Disable Hookify
-
-Target: `~/.claude/settings.json`
-
-```bash
-node -e "
-  const fs = require('fs');
-  const p = require('os').homedir() + '/.claude/settings.json';
-  const s = JSON.parse(fs.readFileSync(p, 'utf8'));
-  const key = Object.keys(s.enabledPlugins || {}).find(k => k.includes('hookify'));
-  if (key && s.enabledPlugins[key] !== false) {
-    s.enabledPlugins[key] = false;
-    fs.writeFileSync(p, JSON.stringify(s, null, 2) + '\n');
-    console.log('Disabled hookify');
-  } else {
-    console.log('Hookify already disabled — skipped');
-  }
-"
-```
-
----
-
 ## M1: Add Read-Only Tools to SKIP_TOOLS
 
 Target: `~/.claude-mem/settings.json`

package/skills/brace/references/report-template.md

@@ -19,6 +19,7 @@ Present this summary after verification completes.
 │
 │  📄 Files
 │     {✅|⏭️} CLAUDE.md
+│     {✅|⏭️} ~/.claude/CLAUDE.md  {updated / already present / skipped (project-only)}
 │     {✅|⏭️} .gitignore
 │
 │  🗑️ Removed (only if legacy items were cleaned up)
@@ -27,7 +28,6 @@ Present this summary after verification completes.
 │     {✅} memory/           Legacy memory directory
 │
 │  🔌 Plugin Tuning (only if Phase 7 ran)
-│     {✅|⏭️} Hookify: {disabled / already disabled / not installed / skipped / N/A}
 │     {✅|⏭️} claude-mem SKIP_TOOLS: {optimised / already optimal / not installed / skipped / N/A}
 │     {✅|⏭️} claude-mem context: {reduced / already optimal / not installed / skipped / N/A}
 │     {✅|⏭️} claude-mem provider: {switched / already optimal / not installed / skipped / N/A}

package/skills/brace/references/scaffold-manifest.md

@@ -9,7 +9,7 @@ creation checklist. Phase 1 uses it to detect existing structure.
 |------|--------|-------------|
 | `CLAUDE.md` | references/claude-md-template.md | Project operating document |
 | `.gitignore` | assets/gitignore-template | Standard ignores |
-| `~/.claude/CLAUDE.md` | assets/global-preferences-template.md | Global preferences (conditional on install_level) |
+| `~/.claude/CLAUDE.md` | assets/global-preferences-template.md | Global preferences (when install_level is "global") |
 
 ## Project Directories
 

package/skills/dock/references/pipeline-templates.md

@@ -115,9 +115,18 @@ jobs:
 
       - name: Smoke test
         run: |
-          # Wait for deployment to be ready, then hit health endpoint
-          sleep 10
-          curl -f "$DEV_URL/healthz" || exit 1
+          # Poll the health endpoint until ready (bounded deadline so a
+          # genuinely broken deploy still fails fast). Do not use a blind
+          # `sleep N && curl` — it flakes on slow cold starts and wastes
+          # time on fast ones.
+          deadline=$((SECONDS + 120))
+          until curl -fs "$DEV_URL/healthz" >/dev/null; do
+            if [ $SECONDS -ge $deadline ]; then
+              echo "Health check failed to pass within 120s" >&2
+              exit 1
+            fi
+            sleep 2
+          done
 
   # ── Promote to Staging (on release tag) ───────────────────────
   promote-staging:
@@ -173,8 +182,15 @@ jobs:
 
       - name: Post-deploy smoke test
         run: |
-          sleep 10
-          curl -f "$PROD_URL/healthz" || exit 1
+          # Poll until healthy, bounded deadline. Never `sleep N && curl`.
+          deadline=$((SECONDS + 120))
+          until curl -fs "$PROD_URL/healthz" >/dev/null; do
+            if [ $SECONDS -ge $deadline ]; then
+              echo "Production health check failed to pass within 120s" >&2
+              exit 1
+            fi
+            sleep 2
+          done
 ```
 
 ### Reusable deploy step patterns

package/skills/launch/SKILL.md

@@ -0,0 +1,355 @@
+---
+name: launch
+description: "Run the full OMC idea-to-merged-PR pipeline — cancel + deep-interview + ralplan + autopilot + mad-skills:ship — in a single invocation. Explicit-only; this skill never auto-activates. Only run when the user literally types /launch. Do not invoke on phrases like \"launch this\", \"ship it\", \"full pipeline\", or similar — none of those should trigger this skill."
+argument-hint: "<rough idea for the feature>"
+allowed-tools: Bash, Read, Skill
+disable-model-invocation: true
+---
+
+# Launch — Idea → Merged PR, end-to-end
+
+When this skill is invoked, IMMEDIATELY output the banner below before doing anything else.
+Pick ONE tagline at random — vary your choice each time.
+CRITICAL: Reproduce the banner EXACTLY character-for-character. The first line of the art has 4 leading characters (one invisible braille-blank + 3 spaces) — you MUST preserve them.
+
+```
+{tagline}
+
+⠀   ██╗██╗      █████╗ ██╗   ██╗███╗   ██╗ ██████╗██╗  ██╗
+   ██╔╝██║     ██╔══██╗██║   ██║████╗  ██║██╔════╝██║  ██║
+  ██╔╝ ██║     ███████║██║   ██║██╔██╗ ██║██║     ███████║
+ ██╔╝  ██║     ██╔══██║██║   ██║██║╚██╗██║██║     ██╔══██║
+██╔╝   ███████╗██║  ██║╚██████╔╝██║ ╚████║╚██████╗██║  ██║
+╚═╝    ╚══════╝╚═╝  ╚═╝ ╚═════╝ ╚═╝  ╚═══╝ ╚═════╝╚═╝  ╚═╝
+```
+
+Taglines:
+- 🚀 From idea to orbit!
+- 🛫 Cleared for takeoff!
+- 🎯 Zero-touch launch inbound!
+- 🎆 3, 2, 1… liftoff!
+- 🛰️ Launch sequence initiated!
+- 🏁 One command, full launch!
+- 🤖 Full auto, full send!
+- ⚙️ The whole launchpad!
+
+---
+
+## Output Formatting
+
+After the banner, display parsed input:
+
+```
+┌─ Input ────────────────────────────────────────
+│  {Field}:  {value}
+│  Flags:    {parsed flags or "none"}
+└────────────────────────────────────────────────
+```
+
+Pre-flight results:
+
+```
+── Pre-flight ───────────────────────────────────
+  ✅ {dep}           {version or "found"}
+  ⚠️ {dep}           not found → {fallback detail}
+  ❌ {dep}           missing → stopping
+──────────────────────────────────────────────────
+```
+
+Stage headers: `━━ {N} · {Name} ━━━━━━━━━━━━━━━━━━━━━━━━━`
+
+Status icons: ✅ done · ❌ failed · ⚠️ degraded · ⏳ working · ⏭️ skipped
+
+---
+
+## Purpose
+
+`/launch` is a thin orchestrator over five existing skills. Its job is to run them in order and keep going between them — nothing more. Each inner skill owns its own stage-level concerns (state, hooks, reviewers, regression). Treat this wrapper as dumb: if any inner stage fails, stop and report — do not attempt retry logic here.
+
+The collaboration window is **Stage 2 (deep-interview)**. Everything after is autonomous; any mid-pipeline pauses come from the inner skills surfacing `AskUserQuestion` prompts (e.g. an autopilot Phase 4 reviewer needs a call), not from this wrapper.
+
+## Flags
+
+Parse optional flags from the request:
+
+- `--skip-cancel`: skip the Stage 1 stale-state cleanup (use when you *know* the session is clean)
+- `--skip-interview`: skip Stage 2 if a usable spec already exists at `.omc/specs/deep-interview-*.md`
+- `--pr-only`: stop after autopilot (Stage 4); don't run ship
+- `--no-ship`: alias for `--pr-only`
+- `--critic=architect|critic|codex`: forwarded to ralplan for the consensus reviewer
+
+---
+
+## Pre-flight
+
+Before Stage 1, check all dependencies in this table:
+
+| Dependency | Type | Check | Required | Resolution | Detail |
+|---|---|---|---|---|---|
+| git | cli | `git --version` | yes | stop | Install from https://git-scm.com |
+| oh-my-claudecode | plugin | `ls "$HOME/.claude/plugins/cache/omc/oh-my-claudecode"/*/skills/cancel/SKILL.md 2>/dev/null` | yes | stop | OMC plugin not installed. Install via `/plugin add oh-my-claudecode` or follow the OMC installation docs. Without it, deep-interview / ralplan / autopilot / cancel are unavailable. |
+| oh-my-claudecode:cancel | skill | `ls "$HOME/.claude/plugins/cache/omc/oh-my-claudecode"/*/skills/cancel/SKILL.md 2>/dev/null` | yes | stop | OMC cancel skill missing — reinstall OMC |
+| oh-my-claudecode:deep-interview | skill | `ls "$HOME/.claude/plugins/cache/omc/oh-my-claudecode"/*/skills/deep-interview/SKILL.md 2>/dev/null` | yes | stop | OMC deep-interview skill missing — reinstall OMC |
+| oh-my-claudecode:ralplan | skill | `ls "$HOME/.claude/plugins/cache/omc/oh-my-claudecode"/*/skills/ralplan/SKILL.md 2>/dev/null` | yes | stop | OMC ralplan skill missing — reinstall OMC |
+| oh-my-claudecode:autopilot | skill | `ls "$HOME/.claude/plugins/cache/omc/oh-my-claudecode"/*/skills/autopilot/SKILL.md 2>/dev/null` | yes | stop | OMC autopilot skill missing — reinstall OMC |
+| mad-skills:ship | skill | `ls "$HOME/.claude/plugins/marketplaces/slamb2k/skills/ship/SKILL.md" 2>/dev/null` | yes | stop | mad-skills:ship not installed. Skip with `--pr-only` to stop after autopilot. Install with: `npx skills add slamb2k/mad-skills --skill ship` |
+
+For each row, in order:
+
+1. Skip rows that don't apply (e.g. the `mad-skills:ship` row when `--pr-only` is set — the skill isn't required in that path)
+2. Run the Check command
+3. If found: continue silently
+4. If missing: apply Resolution strategy
+   - **stop**: notify user with Detail, halt execution
+5. After all checks: summarize what's available and what's degraded
+
+The most important check is the **oh-my-claudecode plugin** — without it, four of the five stages are impossible and the skill must abort. Do not try to proceed with only some OMC skills installed; if the cache directory exists but individual skills are missing, the OMC installation is corrupt and needs reinstalling.
+
+**Additional state check (soft warning, not a blocker):** after the dependency table, run `git status --porcelain` once. If there is any output, warn the user that the working tree has uncommitted changes. `deep-interview` and `ralplan` don't touch source, but `autopilot` will. Let the user decide whether to stash / commit / proceed.
+
+---
+
+## Plan Resolution
+
+Capture the user's argument as the **IDEA** (the rough feature description). This is what deep-interview will use as its starting input.
+
+Display in the Input box as:
+
+```
+│  Idea:   {first 80 chars of idea}
+│  Flags:  {parsed flags or "none"}
+```
+
+---
+
+## Stage 1 · Cancel stale state
+
+Unless `--skip-cancel` was set, invoke the OMC cancel skill to wipe any orphaned state:
+
+```
+Skill("oh-my-claudecode:cancel", "--force")
+```
+
+**Why force:** `/launch` is meant to be a clean-slate starting point. If previous OMC sessions left orphaned `awaiting_confirmation` entries (this is a known rough edge — e.g. the word "autopilot" in a prompt can trigger autopilot state without user consent), those will pollute later stages. `--force` wipes everything so the downstream stages start from zero.
+
+After the cancel skill returns, confirm no modes are active before proceeding. Emit a short one-line status:
+
+```
+  ✅ state cleared — no active OMC modes
+```
+
+If cancel reports lingering state it couldn't clear, stop and report. Don't press on.
+
+---
+
+## Stage 2 · Deep-interview (requirements Q&A)
+
+Unless `--skip-interview` was set AND a pre-existing spec is found, invoke:
+
+```
+Skill("oh-my-claudecode:deep-interview", "{IDEA}")
+```
+
+Let deep-interview run to completion. It will:
+
+- Conduct Socratic Q&A via `AskUserQuestion` — **these pauses surface to the user**, not the wrapper
+- Continue until its ambiguity gate passes (≤ 20%)
+- Write the final spec to `.omc/specs/deep-interview-{slug}.md`
+
+**If `--skip-interview` was set:**
+
+- Look for any file matching `.omc/specs/deep-interview-*.md`
+- If found, use the most recently modified one and note its path in the stage output
+- If not found, warn the user and fall back to running deep-interview anyway
+
+When the stage completes, emit:
+
+```
+  ✅ spec written to .omc/specs/deep-interview-{slug}.md
+```
+
+---
+
+## Stage 3 · Ralplan (plan + consensus review)
+
+Invoke:
+
+```
+Skill("oh-my-claudecode:ralplan", "--direct{critic-flag-if-any}")
+```
+
+`--direct` tells ralplan to skip its own interview and consume the deep-interview spec directly. If the user passed `--critic=architect|critic|codex`, forward that flag to ralplan.
+
+Ralplan will run Planner → Architect → Critic in sequence, looping until Critic approves. When it completes, the consensus plan lives at `.omc/plans/ralplan-{slug}.md`.
+
+Emit:
+
+```
+  ✅ consensus plan at .omc/plans/ralplan-{slug}.md
+  ✅ Critic verdict: APPROVED
+```
+
+If Critic keeps rejecting past its internal loop budget, ralplan will stop and report. In that case, stop this wrapper too — don't press into autopilot with a rejected plan.
+
+---
+
+## Stage 4 · Autopilot (execution + QA + validation)
+
+Invoke:
+
+```
+Skill("oh-my-claudecode:autopilot")
+```
+
+No arguments. Autopilot will detect the consensus plan at `.omc/plans/ralplan-{slug}.md` and **skip its own Phase 0 + Phase 1** (expansion + planning), going straight to Phase 2 (Execution). It runs:
+
+- Phase 2: parallel executor agents implement the plan
+- Phase 3: QA cycling (up to 5 iterations) — build, lint, test, fix
+- Phase 4: multi-perspective validation (Architect + Security-reviewer + Code-reviewer in parallel; all must approve)
+- Phase 5: state cleanup
+
+**Critical:** autopilot's output is *verified local code*, not a merged PR. Autopilot's Phase 5 is only state-file cleanup — it does not push, open a PR, watch CI, or merge. Ship is a separate stage.
+
+Emit:
+
+```
+  ✅ all phases complete
+  ✅ tests pass, build green, validators approved
+```
+
+If autopilot's Phase 3 QA cycle repeats the same error 3 times, or Phase 4 validation fails 3 rounds, autopilot stops and reports. In that case stop this wrapper too — a human needs to review.
+
+---
+
+## Stage 5 · Ship (push → PR → merge → sync)
+
+**Skip if `--pr-only` or `--no-ship` was set.**
+
+Invoke:
+
+```
+Skill("mad-skills:ship", "{one-line summary of approach from ralplan} — Files: {files from autopilot IMPL_REPORT}")
+```
+
+`mad-skills:ship` handles the PR lifecycle:
+
+- Pushes the branch (creates one if needed)
+- Creates PR via `gh pr create` (or AzDO equivalent)
+- Watches CI — fixes failures in a separate sub-agent up to 2 attempts
+- Squash-merges on green
+- Syncs local back to default branch
+
+Emit:
+
+```
+  ✅ PR #{number} merged ({merge_commit})
+  ✅ local synced to {default_branch}
+```
+
+If ship's CI watch exhausts its fix attempts, ship stops and displays its own failure banner. In that case, stop — do not attempt additional retry here.
+
+---
+
+## Waiting patterns
+
+Whenever this skill (or its inner invocations) needs to wait on a file or condition, use a polling loop with an explicit deadline — **never** `sleep N && <cmd>`:
+
+```bash
+deadline=$((SECONDS + 600))
+until [ -s "$path" ] || [ $SECONDS -ge $deadline ]; do sleep 2; done
+```
+
+This keeps waits bounded and cache-friendly, and lets the loop exit early when the expected file appears.
+
+---
+
+## Final Report
+
+```
+┌─ Launch · Report ──────────────────────────────
+│
+│  ✅ Launch complete
+│
+│  💡 Idea:       {first line of idea}
+│  📄 Spec:       .omc/specs/deep-interview-{slug}.md
+│  🗺️ Plan:       .omc/plans/ralplan-{slug}.md
+│
+│  📝 Stages
+│     1. State cleared             ✅
+│     2. Requirements interview    ✅  ({Q&A rounds} rounds)
+│     3. Plan + Critic consensus   ✅  ({critic-approval-rounds} rounds)
+│     4. Execution + QA + review   ✅  ({QA cycles} QA cycles, {validator} approved)
+│     5. PR + merge                ✅  ({pr_url})
+│
+│  📊 Code
+│     Files changed:  {count}
+│     Tests:          {passed}/{total}
+│
+│  🔗 Links
+│     PR:    {pr_url}
+│     Merge: {merge_commit}
+│
+│  ⚡ Next
+│     {anything surfaced by debrief from autopilot or ship}
+│
+└─────────────────────────────────────────────────
+```
+
+If the run stopped short (`--pr-only`, `--no-ship`, or a stage failure), adjust the stages list accordingly — mark later stages with ⏭️ (skipped) or ❌ (failed) and surface what's left to do manually in the `⚡ Next` section.
+
+### Pipeline Summary (always emit)
+
+```
+┌─ Pipeline Summary ─────────────────────────────
+│
+│  {icon} Cancel         {"no-op" or "cleared N sessions"}
+│  {icon} Deep-interview {".omc/specs/... or reused spec"}
+│  {icon} Ralplan        {critic verdict}
+│  {icon} Autopilot      {"green / N files changed"}
+│  {icon} Ship           {pr_url → merge_commit}
+│
+└─────────────────────────────────────────────────
+```
+
+---
+
+## Failure Handling
+
+`/launch` is deliberately dumb about retries. Each inner skill already has its own retry/verification logic — trying to add more at this layer just fights with them. If any stage stops with a failure:
+
+1. Display the failure banner below (not the success report)
+2. List which stages completed, which failed, and which were skipped
+3. Stop. Do not advance to later stages. Do not invoke `/sync` or any cleanup.
+
+```
+┌─ Launch · FAILED ───────────────────────────────
+│
+│  ❌ Pipeline stopped at Stage {N}: {stage name}
+│
+│  Reason:           {specific failure reason from the inner skill}
+│  Stages completed: {list}
+│  Stages skipped:   {list}
+│
+│  ⚠️  Work may exist locally but is NOT merged.
+│     Review the inner skill's output for next steps.
+│
+└──────────────────────────────────────────────────
+```
+
+Rules on failure:
+
+- Do not emit a "Pipeline Summary" with a ✅
+- Do not suggest a cancel/sync/retry here — the inner skill that failed owns that
+- Do not invoke any recovery skill automatically
+
+---
+
+## Trigger Behaviour
+
+This skill has `disable-model-invocation: true`. That means:
+
+- It will **never** auto-activate on keyword hits, even if the user says things like *"launch this feature"*, *"ship it end to end"*, *"run the full pipeline"*, or mentions any of the inner skill names.
+- It runs **only** when the user literally types `/launch` as a slash command.
+- Another skill can still invoke it explicitly via `Skill("launch", "...")` if desired.
+
+This is deliberate: the inner skills (especially autopilot) have aggressive keyword triggers of their own, and a wrapper that also auto-triggered would compound the false-positive risk. Keep this file's `disable-model-invocation` flag set.

package/skills/manifest.json

@@ -1,12 +1,12 @@
 {
-  "generated": "2026-04-17T04:59:26.749Z",
-  "count": 10,
+  "generated": "2026-04-20T02:26:55.864Z",
+  "count": 11,
   "skills": [
     {
       "name": "brace",
       "directory": "brace",
       "description": "'Initialize any project directory with a standard scaffold for AI-assisted development. Creates specs/ and context/ directories, a project CLAUDE.md with development workflow and guardrails, .gitignore, and branch protection. Recommends claude-mem for persistent memory. Idempotent — safe to run on existing projects. Triggers: \"init project\", \"setup brace\", \"brace\", \"initialize\", \"bootstrap\", \"scaffold\".'",
-      "lines": 427,
+      "lines": 424,
       "hasScripts": false,
       "hasReferences": true,
       "hasAssets": true,
@@ -52,6 +52,16 @@
       "hasAssets": false,
       "hasTests": true
     },
+    {
+      "name": "launch",
+      "directory": "launch",
+      "description": "\"Run the full OMC idea-to-merged-PR pipeline — cancel + deep-interview + ralplan + autopilot + mad-skills:ship — in a single invocation. Explicit-only; this skill never auto-activates. Only run when the user literally types /launch. Do not invoke on phrases like \\\"launch this\\\", \\\"ship it\\\", \\\"full pipeline\\\", or similar — none of those should trigger this skill.\"",
+      "lines": 356,
+      "hasScripts": false,
+      "hasReferences": false,
+      "hasAssets": false,
+      "hasTests": false
+    },
     {
       "name": "prime",
       "directory": "prime",

package/skills/ship/scripts/merge.sh

@@ -94,30 +94,32 @@ if [ "$AZDO_MODE" = "cli" ]; then
     sleep 15
   done
 
-  # Complete the PR
-  if az repos pr update --id "$PR_NUMBER" --status completed \
-    --org "$AZDO_ORG_URL" \
-    --squash "$SQUASH_FLAG" \
-    --delete-source-branch "$DELETE_FLAG" 2>/dev/null; then
-    STATUS="success"
-    MERGE_COMMIT=$(git rev-parse --short HEAD 2>/dev/null)
-    BRANCH_DELETED=$DELETE_BRANCH
-  else
-    # Retry once after 30s (policies may still be evaluating)
-    sleep 30
+  # Complete the PR — retry on a bounded deadline; policies may still be
+  # evaluating after the initial poll finished. First attempt runs
+  # immediately, then we poll every 5s up to a 30s total budget.
+  # Avoid `sleep 30 && retry` (blind wait) — this loop exits early on success.
+  MERGE_DEADLINE=$((SECONDS + 30))
+  MERGE_OK=false
+  while :; do
     if az repos pr update --id "$PR_NUMBER" --status completed \
       --org "$AZDO_ORG_URL" \
       --squash "$SQUASH_FLAG" \
       --delete-source-branch "$DELETE_FLAG" 2>/dev/null; then
-      STATUS="success"
-      MERGE_COMMIT=$(git rev-parse --short HEAD 2>/dev/null)
-      BRANCH_DELETED=$DELETE_BRANCH
-    else
-      STATUS="failed"
-      ERRORS="Merge failed after retry"
-      MERGE_COMMIT=""; BRANCH_DELETED=false
-      emit_report; exit 2
+      MERGE_OK=true
+      break
     fi
+    [ $SECONDS -ge $MERGE_DEADLINE ] && break
+    sleep 5
+  done
+  if $MERGE_OK; then
+    STATUS="success"
+    MERGE_COMMIT=$(git rev-parse --short HEAD 2>/dev/null)
+    BRANCH_DELETED=$DELETE_BRANCH
+  else
+    STATUS="failed"
+    ERRORS="Merge failed after retry"
+    MERGE_COMMIT=""; BRANCH_DELETED=false
+    emit_report; exit 2
   fi
   emit_report
   exit 0