@lifeaitools/rdc-skills 0.25.0 → 0.25.1

package/.claude-plugin/plugin.json

@@ -1334,6 +1334,45 @@
       "enabled_default": true,
       "codeflow_required": false
     },
+    "env": {
+      "name": "env",
+      "slash": "rdc:env",
+      "category": "infra",
+      "usage": "rdc:env [status|install|repair|update]",
+      "args": {
+        "positional": [
+          {
+            "name": "action",
+            "required": false,
+            "default": "status",
+            "values": ["status", "install", "repair", "update"]
+          }
+        ],
+        "flags": []
+      },
+      "requires": [
+        "clauth"
+      ],
+      "produces": [],
+      "default_model": "sonnet",
+      "triggers": [
+        "check environment",
+        "install environment",
+        "repair environment",
+        "update environment",
+        "setup env",
+        "fix env",
+        "env status",
+        "after reboot",
+        "GPU crash"
+      ],
+      "follows": [],
+      "leads_to": ["status"],
+      "sandbox_aware": false,
+      "output_contract": "guides/output-contract.md",
+      "enabled_default": true,
+      "codeflow_required": false
+    },
     "terminal-config": {
       "name": "terminal-config",
       "slash": "rdc:terminal-config",
@@ -1436,7 +1475,7 @@
       "name": "onramp",
       "slash": "rdc:onramp",
       "category": "infra",
-      "usage": "rdc:onramp <slug> --name \"<Display Name>\" [--location <json>] [--archetype <archetype>] [--owner place-fund|rdc|jv|client] [--history] [--dry-run]",
+      "usage": "rdc:onramp <slug> --name \"<Display Name>\" [--location <json>] [--archetype <archetype>] [--owner place-fund|rdc|jv|client] [--history] [--skip-to <phase>] [--no-gate] [--dry-run]",
       "args": {
         "positional": [
           {
@@ -1494,7 +1533,10 @@
       ],
       "produces": [
         "work_items",
-        "supabase_writes"
+        "supabase_writes",
+        "git_commits",
+        "git_pushes",
+        "deploys"
       ],
       "default_model": "opus",
       "triggers": [
@@ -1503,12 +1545,12 @@
         "lay the infra for",
         "enroll this project",
         "start a new place fund project",
-        "run enrollment for"
+        "run enrollment for",
+        "run the full pipeline for",
+        "build a place from scratch"
       ],
       "follows": [],
-      "leads_to": [
-        "design"
-      ],
+      "leads_to": [],
       "sandbox_aware": true,
       "output_contract": "guides/output-contract.md",
       "enabled_default": true,

package/CHANGELOG.md

@@ -7,6 +7,18 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ---
 
+## v0.25.0 — rdc:onramp Phase-1 enrollment skill
+
+### Added
+- `rdc:onramp` — Phase-1 enrollment infrastructure orchestrator for new Place
+  Fund projects. It gates the existing `enroll_place` RPC, `insert_work_item`
+  enrollment epic, and `onramp-scaffold-place.mjs` disk scaffolder in strict
+  order, with fresh/already-enrolled/spine-ahead/disk-ahead drift handling.
+- Tier-2 onramp test manifest coverage for the `_context.md` absence invariant,
+  deterministic door contracts, and four-state drift behavior.
+
+---
+
 ## v0.21.0 — channel-formatter skill
 
 ### Added

package/MANIFEST.md

@@ -3,7 +3,7 @@
 **Created:** 2026-04-13  
 **Last updated:** 2026-05-01  
 **Source:** Regen Root project `.claude/skills/user/`  
-**Files:** 18+ skill files, 10 agent guides, command mirrors, and support scripts  
+**Files:** 31 skill files, 10 agent guides, command mirrors, and support scripts  
 **Size:** ~480K  
 
 ## Generic Conversion Summary
@@ -54,6 +54,7 @@ All skills have been converted from project-specific to generic/portable:
     ├── workitems/SKILL.md
     ├── design/SKILL.md       [RDC/Studio design, Palette Library, Rampa CLI]
     ├── help/SKILL.md
+    ├── regen-media/SKILL.md [primary image generation + Regen Media upload]
     ├── deploy/SKILL.md       [Coolify deployment ops]
     ├── release/SKILL.md      [atomic version release]
     ├── self-test/SKILL.md    [static lint + tier2 smoke tests]
@@ -113,6 +114,8 @@ All skills have been converted from project-specific to generic/portable:
 - rdc:channel-formatter — channel-native formatting and content repurposing,
   including article-to-post, social-pack, campaign-pack, exec-pack, and
   launch-pack outputs
+- rdc:regen-media — primary image generation and Regen Media upload workflow,
+  defaulting to local keyless Codex `gpt-image-2` via built-in `image_gen`
 
 ### Bridge (3)
 - rdc:handoff — planning → work items

package/README.md

@@ -13,7 +13,7 @@ From any Claude Code session:
 /plugin install rdc-skills
 ```
 
-The full catalog is available through the MCP as 30 skills. Nineteen common
+The full catalog is available through the MCP as 31 skills. Nineteen common
 workflows also ship as `/rdc:*` command shorthands. Agent playbooks in
 `guides/agents/` are dispatched internally by `rdc:build` and are not
 user-invocable.
@@ -181,7 +181,7 @@ Throughout the skills, the following substitutions have been made to make them p
 
 ## Skills Overview
 
-30 MCP skills organized into 8 manifest categories. Nineteen of these also have
+31 MCP skills organized into 8 manifest categories. Nineteen of these also have
 `/rdc:*` command shorthands for Claude Code/plugin callers.
 
 Use `rdc_skill_list` for the authoritative live catalog, including names,
@@ -216,6 +216,7 @@ Located in `guides/agents/` — plain markdown playbooks spawned as sub-agents:
 - lifeai-brochure-author — JSX authoring contract for Brochurify inputs
 - rdc:convert — Office/Markdown conversion via build-corpus (work in Markdown, materialize `.docx`/`.pptx` at the end; round-trips losslessly — see build-corpus `AGENTS.md`)
 - rdc:fs-mcp — File System MCP bridge guidance for live repo reads/writes
+- rdc:regen-media — primary image generation and Regen Media upload workflow; defaults to local keyless Codex `gpt-image-2` via built-in `image_gen`
 - rdc:extract-verifier-rules — verifier-rule extraction from enhancement logs
 - rdc:rpms-filemap — generated RPMS file-map guidance
 - rdc:workitems — manual work item CRUD
@@ -313,6 +314,7 @@ skills/
   edit/SKILL.md        (local website editor launcher)
   fixit/SKILL.md       (quick-fix bypass)
   fs-mcp/SKILL.md      (File System MCP bridge guidance)
+  regen-media/SKILL.md (primary image generation + Regen Media upload)
   handoff/SKILL.md     (planning → work items)
   help/SKILL.md        (manifest-driven skill index)
   housekeeping/SKILL.md (weekly maintenance audit)

package/commands/overnight.md

@@ -73,7 +73,7 @@ Determine which epics to work:
   - `status IN ('todo', 'in_progress')`
   - Order: urgent first, then high, then by `created_at`
 
-Log the epic queue at the start of `.rdc/reports/overnight-<YYYY-MM-DD>.md` (fallback: `.rdc/reports/overnight-<YYYY-MM-DD>.md` if `.rdc/` does not exist).
+Log the epic queue at the start of `.rdc/overnights/overnight-<YYYY-MM-DD>.md` (fallback: `.rdc/overnights/overnight-<YYYY-MM-DD>.md` if `.rdc/` does not exist).
 
 ## Phase 3 — Epic Loop
 
@@ -145,7 +145,7 @@ After all epics are processed:
 
 1. Run `rdc:report --unattended`
 
-2. Write session summary to `.rdc/reports/overnight-<YYYY-MM-DD>.md` (fallback: `.rdc/reports/overnight-<YYYY-MM-DD>.md`):
+2. Write session summary to `.rdc/overnights/overnight-<YYYY-MM-DD>.md` (fallback: `.rdc/overnights/overnight-<YYYY-MM-DD>.md`):
    ```markdown
    # Overnight Session — YYYY-MM-DD
 

package/git-sha.json

@@ -1,3 +1,3 @@
 {
-  "sha": "d6605a5c16373e540f31b046c1a5687170893d0e"
+  "sha": "bf46e3a7c1b60f99831e4e087c53007083eac436"
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lifeaitools/rdc-skills",
-  "version": "0.25.0",
+  "version": "0.25.1",
   "description": "RDC typed-agent dispatch skill suite for Claude Code - plan, build, review, overnight builds",
   "keywords": [
     "claude-code",

package/scripts/install-rdc-skills.js

@@ -30,6 +30,8 @@ const fs           = require('fs');
 const path         = require('path');
 const os           = require('os');
 const readline     = require('readline');
+const http         = require('http');
+const crypto       = require('crypto');
 const { execSync } = require('child_process');
 
 // ── Args ──────────────────────────────────────────────────────────────────────
@@ -102,6 +104,12 @@ function run(cmd, options = {}) {
   return execSync(cmd, { encoding: 'utf8', stdio: 'pipe', ...options }).trim();
 }
 
+function shellQuote(s) {
+  const raw = String(s);
+  if (process.platform === 'win32') return `"${raw.replace(/"/g, '\\"')}"`;
+  return `'${raw.replace(/'/g, `'\\''`)}'`;
+}
+
 function updateCodexMcpToml(toml, mcpUrl) {
   const blockRe = /(^|\n)(\[mcp_servers\.rdc-skills\]\n)([\s\S]*?)(?=\n\[|\s*$)/;
   const desiredLine = `url = '${mcpUrl}'`;
@@ -322,6 +330,35 @@ function readInstalledPackageVersion(scriptPath) {
   return readJson(path.join(packageRoot, 'package.json'), {}).version || null;
 }
 
+function collectFilesRecursive(baseDir, rel, out) {
+  const full = path.join(baseDir, rel);
+  if (!fs.existsSync(full)) return;
+  const stat = fs.statSync(full);
+  if (stat.isDirectory()) {
+    for (const entry of fs.readdirSync(full).sort()) {
+      collectFilesRecursive(baseDir, path.join(rel, entry), out);
+    }
+    return;
+  }
+  if (stat.isFile()) out.push(rel.replace(/\\/g, '/'));
+}
+
+function hashInstallSurface(root) {
+  if (!root || !fs.existsSync(root)) return null;
+  const rels = [];
+  for (const rel of ['.claude-plugin', 'commands', 'guides', 'skills', 'package.json', 'README.md']) {
+    collectFilesRecursive(root, rel, rels);
+  }
+  const hash = crypto.createHash('sha256');
+  for (const rel of rels.sort()) {
+    hash.update(rel);
+    hash.update('\0');
+    hash.update(fs.readFileSync(path.join(root, rel)));
+    hash.update('\0');
+  }
+  return hash.digest('hex');
+}
+
 function syncGlobalMcpInstall(version) {
   const proc = getPm2Process(MCP_NAME);
   if (!proc) {
@@ -336,18 +373,22 @@ function syncGlobalMcpInstall(version) {
   }
 
   const installedVersion = readInstalledPackageVersion(scriptPath);
-  if (installedVersion === version) {
-    ok(`[2.9] MCP pkg    — global ${NPM_PACKAGE}@${version} already installed`);
+  const packageRoot = packageRootFromMcpScript(scriptPath);
+  const sourceHash = hashInstallSurface(repoRoot);
+  const installedHash = hashInstallSurface(packageRoot);
+  if (installedVersion === version && sourceHash && installedHash && sourceHash === installedHash) {
+    ok(`[2.9] MCP pkg    — global ${NPM_PACKAGE}@${version} already matches source`);
     return;
   }
 
   let stopped = false;
   let installed = false;
   try {
-    info(`[2.9] MCP pkg    — updating global ${NPM_PACKAGE} ${installedVersion || '?'} → ${version}`);
+    const reason = installedVersion === version ? 'same version, source changed' : `${installedVersion || '?'} → ${version}`;
+    info(`[2.9] MCP pkg    — updating global ${NPM_PACKAGE} (${reason}) from local source`);
     run(`pm2 stop ${MCP_NAME}`);
     stopped = true;
-    run(`npm install -g ${NPM_PACKAGE}@${version}`);
+    run(`npm install -g ${shellQuote(repoRoot)}`);
     installed = true;
     ok(`[2.9] MCP pkg    — installed global ${NPM_PACKAGE}@${version}`);
   } catch (e) {
@@ -367,6 +408,71 @@ function syncGlobalMcpInstall(version) {
   }
 }
 
+function callLocalMcpTool(name, args = {}) {
+  return new Promise((resolve, reject) => {
+    const payload = JSON.stringify({
+      jsonrpc: '2.0',
+      id: 1,
+      method: 'tools/call',
+      params: { name, arguments: args },
+    });
+    const req = http.request({
+      hostname: '127.0.0.1',
+      port: Number(MCP_PORT),
+      path: '/mcp',
+      method: 'POST',
+      headers: {
+        'Content-Type': 'application/json',
+        'Accept': 'application/json, text/event-stream',
+        'Content-Length': Buffer.byteLength(payload),
+      },
+      timeout: 10000,
+    }, (res) => {
+      let body = '';
+      res.setEncoding('utf8');
+      res.on('data', (chunk) => { body += chunk; });
+      res.on('end', () => {
+        try {
+          const dataLine = body.split(/\r?\n/).find((line) => line.startsWith('data: '));
+          const envelope = JSON.parse(dataLine ? dataLine.slice(6) : body);
+          const text = envelope?.result?.content?.[0]?.text;
+          resolve(text ? JSON.parse(text) : envelope);
+        } catch (e) {
+          reject(new Error(`invalid MCP response: ${e.message}`));
+        }
+      });
+    });
+    req.on('timeout', () => req.destroy(new Error('MCP request timed out')));
+    req.on('error', reject);
+    req.write(payload);
+    req.end();
+  });
+}
+
+async function verifyLiveMcpCatalogFresh() {
+  const proc = getPm2Process(MCP_NAME);
+  if (!proc) {
+    warn('[7.5] MCP catalog — skipped (PM2 process not registered)');
+    return;
+  }
+  const sourceSkills = fs.readdirSync(path.join(repoRoot, 'skills'), { withFileTypes: true })
+    .filter((entry) => entry.isDirectory() && fs.existsSync(path.join(repoRoot, 'skills', entry.name, 'SKILL.md')))
+    .map((entry) => entry.name)
+    .sort();
+  const catalog = await callLocalMcpTool('rdc_skill_list', {});
+  const rows = Array.isArray(catalog)
+    ? catalog
+    : Array.isArray(catalog?.skills) ? catalog.skills
+    : Array.isArray(catalog?.results) ? catalog.results
+    : [];
+  const liveNames = new Set(rows.map((row) => row.name));
+  const missing = sourceSkills.filter((name) => !liveNames.has(name));
+  if (missing.length > 0 || rows.length < sourceSkills.length) {
+    throw new Error(`live MCP catalog stale: source=${sourceSkills.length}, live=${rows.length}, missing=${missing.join(', ') || '(count mismatch)'}`);
+  }
+  ok(`[7.5] MCP catalog — ${rows.length} live skill(s), source catalog fresh`);
+}
+
 // ── User-skills cleanup ───────────────────────────────────────────────────────
 // Older installer versions wrote skill files directly to ~/.claude/skills/user/.
 // Claude Code loads that directory AND the plugin cache, so any rdc skills left
@@ -1260,6 +1366,12 @@ async function main() {
   console.log('');
   console.log('  \x1b[36mMCP server:\x1b[0m');
   try { registerMcpServer(); } catch (e) { warn(`[7/7] MCP server — unexpected error (${e.message})`); }
+  try {
+    await verifyLiveMcpCatalogFresh();
+  } catch (e) {
+    fail(`[7.5] MCP catalog — ${e.message}`);
+    process.exit(2);
+  }
 
   // Done
   console.log('');

package/skills/build/SKILL.md

@@ -425,6 +425,17 @@ Read the task title and description, then:
      and run it green before the item leaves `review`. Agent prompt line: *"This
      task is not done until a committed test in the same commit exercises every
      surface; for collections, loop all items and assert output == source."*
+  - **Durable workflow E2E gate (hard):** A task that creates a queue/job,
+    background worker, storage hand-off, retry/lease, webhook, or other async
+    cross-process workflow MUST ship a committed `test:e2e` (or equivalent)
+    harness. Before the task leaves `review`, run it against the deploy-equivalent
+    container plus real disposable dependencies (or an isolated mirror) and
+    retain its receipt. The harness MUST exercise the public ingress, assert the
+    full state transition, assert every persisted output, and clean up fixture
+    rows/objects in `finally`; unit tests, mocks, a migration check, Docker
+    construction, or a `/health` probe cannot substitute. If no safe disposable
+    environment exists, create one within the work package; do not waive this
+    gate or mark the item reviewable.
 
 8. **Post-wave test gate (mandatory):**
    After all agents in a wave complete, before proceeding:
@@ -514,6 +525,10 @@ Read the task title and description, then:
     - Runs `npx tsc --noEmit` for every touched app/package
     - Starts the dev server and probes every modified route (expects HTTP 200, not 500)
     - Runs vitest for every touched package
+    - For every durable workflow, runs the committed E2E harness and rejects the
+      item unless its receipt proves public ingress -> queued/running -> terminal
+      state, persisted outputs, and fixture cleanup on the deploy-equivalent
+      runtime; an absent harness is a hard rejection.
     - **Verifies checklist decomposition quality per work item before functional validation:**
       - Every implementation work item has >= 10 attested `decomp-*`/`test-*` items, meets the per-surface completeness floors (each declared surface covered), and no feature WP ships a 5-6-row checklist
       - Every `decomp-*` item includes route/file, action, expected result, and evidence artifact

package/skills/env/SKILL.md

@@ -0,0 +1,141 @@
+---
+name: rdc:env
+description: "Usage `rdc:env [status|install|repair|update]` — Manage the LIFEAI environment harness: check status, install on a fresh box, repair broken services, or pull updates. Reads $LIFEAI_ENV/manifest.json as the source of truth. Use when: 'check the environment', 'install environment', 'repair environment', 'update environment', 'setup env', 'fix env', 'env status', or after a reboot/GPU crash."
+---
+
+> **⚠️ OUTPUT CONTRACT (READ FIRST):** `guides/output-contract.md`
+> Checklist-only output. No tool-call narration. No raw MCP/JSON/log dumps.
+> One checklist upfront, updated in place, shown again at end with a 1-line verdict.
+
+# rdc:env — Environment Harness Manager
+
+## When to Use
+- After a reboot or GPU crash — verify everything came back up
+- Setting up a fresh box — install all tools and services
+- Something is broken — repair services, MCPs, CodeFlow
+- Routine update — pull latest env repo and run drift check
+- "check the environment", "fix my setup", "install environment"
+
+## Subcommands
+
+| Command | What it does |
+|---------|-------------|
+| `status` (default) | Read-only check: env repo version, tool versions, MCP health, shim integrity, CodeFlow brain |
+| `install` | Full provision: clone env repo if missing, set LIFEAI_ENV, install tools, configure MCPs |
+| `repair` | Diagnose and fix: restart crashed services, rebuild stale dists, fix broken shims |
+| `update` | Pull latest env repo, re-run audit, report drift, refresh shims if new scripts added |
+
+## Procedure
+
+### Step 0: Resolve environment root
+
+```bash
+LIFEAI_ENV="${LIFEAI_ENV:-C:/Dev/lifeai-env}"
+```
+
+If `$LIFEAI_ENV` is not set AND the default path doesn't exist:
+- For `install`: clone the repo and set the env var
+- For all others: STOP with `BLOCKED: environment repo not found. Run: rdc:env install`
+
+Read `$LIFEAI_ENV/manifest.json` — this is the harness inventory.
+
+### Step 1: Environment repo state
+
+```bash
+git -C "$LIFEAI_ENV" fetch origin 2>/dev/null
+git -C "$LIFEAI_ENV" rev-list --count HEAD..origin/main
+```
+
+Report: version from manifest.json, commits behind origin, last pull date.
+
+For `update`: pull if behind. For `repair`: pull if behind (stale harness may be the cause).
+
+### Step 2: Tool versions (audit)
+
+Run the audit script:
+```bash
+pwsh.exe -NoProfile -ExecutionPolicy Bypass -File "$LIFEAI_ENV/audit/audit.ps1" -ProjectRoot "$PROJECT_ROOT"
+```
+
+Or read `$PROJECT_ROOT/environment.lock.json` and check each tool:
+
+| Tool | Check | Min Version |
+|------|-------|-------------|
+| node | `node --version` | 22.0.0 |
+| pnpm | `pnpm --version` | 10.0.0 |
+| pm2 | `pm2 --version` | 5.0.0 |
+| clauth | `curl -s http://127.0.0.1:52437/ping` | 1.30.0 |
+| rdc-skills | `npm list -g @lifeai/rdc-skills` | 0.25.0 |
+| gh | `gh --version` | 2.0.0 |
+
+For `install` and `repair`: install or upgrade any tool below min_version using the `install` command from the lock file. This includes rdc-skills itself — the environment repo is the orchestrator that keeps all tools current, including its own skill set. A `repair` or `install` that bumps rdc-skills will also restart the rdc-skills-mcp PM2 process so the new version is live immediately.
+
+### Step 3: Service health
+
+Check each service:
+
+| Service | Health Check | Repair |
+|---------|-------------|--------|
+| clauth daemon | `curl -s http://127.0.0.1:52437/ping` | `$LIFEAI_ENV/services/restart-clauth.bat` |
+| CodeFlow gateway | `curl -s http://127.0.0.1:3109/health` | `node $PROJECT_ROOT/scripts/codeflow-up.mjs` |
+| CodeFlow brain | `/health` → `health.state` should be `operational` | `node $PROJECT_ROOT/scripts/codeflow-up.mjs --restart` |
+| rdc-skills MCP | `pm2 list` shows rdc-skills-mcp online | `pm2 restart rdc-skills-mcp` |
+| PM2 daemon | `pm2 ping` | `pm2 resurrect` |
+| Docker | `docker info` | Start Docker Desktop |
+| Neo4j | `docker inspect codeflow-neo4j` | `docker compose -f $LIFEAI_ENV/services/codeflow/docker-compose.yml up -d` |
+
+For `status`: report only. For `repair`: fix each failing service in order (clauth first, then CodeFlow, then MCPs).
+
+### Step 4: MCP server verification
+
+Run the MCP checker:
+```bash
+pwsh.exe -NoProfile -ExecutionPolicy Bypass -File "$LIFEAI_ENV/services/install-mcps.ps1" -ProjectRoot "$PROJECT_ROOT"
+```
+
+For `repair`: add `-Fix` flag to auto-start missing local MCPs.
+
+### Step 5: Shim integrity
+
+Check that all monorepo shims point at valid targets:
+```bash
+pwsh.exe -NoProfile -ExecutionPolicy Bypass -File "$LIFEAI_ENV/provision.ps1" -ProjectRoot "$PROJECT_ROOT" -SkipTools -DryRun
+```
+
+Look for "broken shim" warnings. For `repair`: re-run provision.ps1 live.
+
+### Step 6: Agent readiness (final gate)
+
+```bash
+pnpm --filter @regen/codeflow startup:readiness
+```
+
+This is the same check the startup guard runs. Zero blockers = environment healthy.
+
+## Output Format
+
+```
+## rdc:env <command> — Environment Health
+
+| Check | Status | Detail |
+|-------|--------|--------|
+| Env repo | ✅ | v0.2.0, 0 behind origin |
+| LIFEAI_ENV | ✅ | C:/Dev/lifeai-env (Machine scope) |
+| Node | ✅ | v22.14.0 (min 22.0.0) |
+| pnpm | ✅ | 10.12.1 (min 10.0.0) |
+| clauth | ✅ | v1.30.2, unlocked |
+| CodeFlow | ✅ | v0.33.11, brain=dev, operational |
+| rdc-skills MCP | ✅ | v0.25.0, online |
+| Shims | ✅ | 47/47 valid |
+| Agent readiness | ✅ | 0 blockers |
+
+**Verdict: environment healthy (9/9 checks pass)**
+```
+
+For `repair`, append:
+```
+## Repairs Applied
+- [x] Restarted clauth daemon
+- [x] Rebuilt CodeFlow dist + restarted gateway
+- [ ] rdc-skills: already running
+```

package/skills/fs-mcp/SKILL.md

@@ -123,7 +123,42 @@ fs_import_git_files remote="origin" ref="claude-ai/docs-upload-123" paths=["docs
 
 This tool must fetch only, restore only named paths, optionally commit only those paths, and never push.
 
-### 5. Safety rules
+### 5. `fs_exec` — always use `shell: false` (the default)
+
+`fs_exec` supports an allowlisted set of commands: `git`, `pnpm`, `npx`,
+`node`, `python3`, `rclone`, `bash`, `cat`, `grep`, `find`, `wc`,
+`sha256sum`, `tsc`, `eslint`.
+
+**Always use `shell: false`** (the default). Pass commands as arrays:
+
+```text
+fs_exec command=["git", "add", ".rdc/plans/my-file.md"]
+fs_exec command=["git", "status", "--porcelain"]
+fs_exec command=["git", "diff", "--cached", "--name-only"]
+fs_exec command=["git", "commit", "-m", "docs(plans): add my-file"]
+```
+
+**Never set `shell: true`** unless you specifically need pipes/redirects.
+`shell: true` spawns the host shell (`pwsh.exe` on Windows), which may not
+be resolvable in the FS MCP server's process context (Store-installed
+PowerShell uses a `WindowsApps` shim path that breaks across process
+contexts). `shell: false` spawns the command directly — `git`, `node`, etc.
+are on stable PATH locations and resolve reliably.
+
+**On `fs_exec` failure with `spawn pwsh.exe ENOENT`:** this means the host
+lacks a stable PowerShell 7 install. Fix: install PowerShell 7 via MSI
+(`winget install --id Microsoft.PowerShell --scope machine --force`). But
+the immediate workaround is always `shell: false` — the allowlisted commands
+don't need a shell to execute.
+
+**Before writing governed files** (`.rdc/plans/`, `.claude/rules/`, any path
+with MDK validation hooks): read an existing exemplar in the same directory
+via `fs_read` to learn the required frontmatter schema. Write complete,
+correct content on the FIRST `fs_write`. A two-stage write creates a stale
+index blob; use `fs_exec command=["git", "add", "<path>"]` to re-stage if
+needed.
+
+### 6. Safety rules
 
 - Never run or request `git pull` for the dirty monorepo.
 - Never checkout a whole branch into the local worktree.
@@ -134,7 +169,7 @@ This tool must fetch only, restore only named paths, optionally commit only thos
 - Stage only imported paths when committing.
 - Never push from FS import unless the user explicitly asks for a push-capable workflow.
 
-### 6. Completion report
+### 7. Completion report
 
 Report:
 

package/skills/new-model/SKILL.md

@@ -0,0 +1,49 @@
+---
+name: new-model
+description: >
+  Register a new Reference Model completely — the yes/no gate that prevents half-registered models.
+  A reference model is DEFINED as registered in all four homes: (1) RPMS project_nodes row
+  (node_type=reference_model), (2) apps/vlas route group /models/<slug>, (3) apps/vlas
+  src/data/models/<slug> definitions, (4) global-corpus Drive directory. If ANY of the four is
+  missing, the model does not exist yet — run this skill.
+  TRIGGER: 'rdc:new-model', 'rdc new model', 'create model', 'new reference model', 'add a model',
+  'register model', 'new model for', 'model for conservation easements', 'is this model registered',
+  'model registration check'.
+---
+
+# rdc:new-model — Reference Model Registration (complete or not at all)
+
+## When to run
+- Creating any new reference model (conservation easement, regen data center, etc.)
+- ANY time a model page/idea exists but you cannot confirm all four homes below
+- Asked "is <model> registered?" → run the CHECK phase, report yes/no per home, offer to complete
+
+## The four homes (a model exists ONLY when all four do)
+1. **RPMS node** — `project_nodes` row: `node_type='reference_model'`, slug matches route slug exactly, status active, non-null description, metadata JSONB `{model_kind, capitals_profile, pathway_version, catalog_version, tabs}` when known.
+2. **Routes** — `apps/vlas/src/app/models/<slug>/` (or `(library)/<slug>/` for library-page models) + entry on the models library index.
+3. **Definitions** — `apps/vlas/src/data/models/<slug>/` (engine/constants/prompts as applicable; may start minimal).
+4. **Corpus** — Google Drive `global-corpus/VLAS/models/<slug>/` with subdirs `_working, _archive, _originals, diagrams, docs`. Drive models root folder ID: `1ua4zXohgZeYsZZTb1kUdnPqfN4eQAHuU`; copy structure from `_template` (`1ZVeq7iycg-dzG_iAKXrbTqIFYa6E26bQ`). Access Drive ONLY via the Google Drive connector (claude.ai) — H:\My Drive is NOT under any FS mount. From the CLI, the local sync path `H:\My Drive\global-corpus\VLAS\models\` may be used directly.
+
+## Procedure
+### Phase 0 — CHECK (always first; idempotent)
+For the given slug, verify each home. Output a 4-line yes/no table. If all yes → report "registered" and stop.
+### Phase 1 — REGISTER (only the missing homes; never duplicate)
+- RPMS: INSERT project_nodes with ON CONFLICT (slug) DO NOTHING; RETURNING id. Slug is kebab-case, matches route.
+- Routes/definitions: scaffold minimal `/models/<slug>/page.tsx` + `src/data/models/<slug>/index.ts` if absent (full build is a separate plan/epic — this skill registers, it does not build the site).
+- Corpus: create `<slug>/` + the five subdirs under the models root.
+### Phase 2 — RECORD
+- Add the model to the library index page if missing.
+- If created from claude.ai: note node id + Drive folder id in the session; if a related epic exists, add a note via update_work_item_status.
+- Commit any repo changes on develop with a Work-Item trailer.
+
+## Rules
+- NEVER create a model in fewer than four homes. Partial registration is the failure mode this skill exists to kill (it happened twice: regen-data-center shipped with a bare node; regenerative-development-and-design shipped with no node and no corpus dir).
+- Slug is identical everywhere. No spaces, no case variants.
+- Models NEVER live on place.fund, PRT, TPF, or RDC surfaces — VLAS only (models are teaching instruments of the standard).
+- Catalog scoping: if the model needs interventions, decide model_slug scoping on stewardship_interventions BEFORE seeding (see library-conventions epic ba1b5d5d).
+
+## Done means
+- [ ] Four-home check table all YES
+- [ ] Node id + Drive folder id reported
+- [ ] Library index lists the model
+- [ ] Repo changes committed (develop, trailer)