ideabox 1.0.0 → 1.0.2

package/.claude-plugin/plugin.json

@@ -0,0 +1,20 @@
+{
+  "name": "ideabox",
+  "version": "1.0.2",
+  "description": "Data-driven project idea engine. Researches real market demand, scores ideas on monetization and open-source impact, and orchestrates a full pipeline from idea to shipped code.",
+  "author": {
+    "name": "Pawan Paudel"
+  },
+  "repository": "https://github.com/pawanpaudel93/ideabox",
+  "homepage": "https://github.com/pawanpaudel93/ideabox",
+  "license": "MIT",
+  "keywords": [
+    "ideabox",
+    "ideas",
+    "project-ideas",
+    "monetization",
+    "research",
+    "developer-tools",
+    "agent-skill"
+  ]
+}

package/README.md

@@ -1,24 +1,19 @@
-<p align="center">
-  <h1 align="center">IdeaBox</h1>
-  <p align="center">Data-driven project idea engine for coding agents</p>
-</p>
+# IdeaBox
 
-<p align="center">
-  <a href="https://www.npmjs.com/package/ideabox"><img src="https://img.shields.io/npm/v/ideabox.svg" alt="npm version"></a>
-  <a href="https://opensource.org/licenses/MIT"><img src="https://img.shields.io/badge/License-MIT-blue.svg" alt="License: MIT"></a>
-  <a href="https://github.com/pawanpaudel93/ideabox"><img src="https://img.shields.io/github/stars/pawanpaudel93/ideabox?style=social" alt="GitHub stars"></a>
-</p>
+[![npm version](https://img.shields.io/npm/v/ideabox.svg)](https://www.npmjs.com/package/ideabox)
+[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://opensource.org/licenses/MIT)
+[![GitHub stars](https://img.shields.io/github/stars/pawanpaudel93/ideabox?style=social)](https://github.com/pawanpaudel93/ideabox)
 
----
-
-Researches real market demand, scores ideas on monetization and open-source impact, and orchestrates a full **9-phase pipeline** from idea to shipped code. Self-improving — gets smarter with every session.
+> Data-driven project idea engine for coding agents. Researches real market demand, scores ideas on monetization and open-source impact, and orchestrates a full **9-phase pipeline** from idea to shipped code. Self-improving — gets smarter with every session.
 
 ## Table of Contents
 
-- [The Problem](#the-problem)
-- [The Solution](#the-solution)
+- [Problem](#the-problem)
+- [Solution](#the-solution)
 - [Features](#features)
 - [Installation](#installation)
+- [Requirements](#requirements)
+- [Permissions](#permissions)
 - [Usage](#usage)
 - [How It Works](#how-it-works)
 - [Scoring](#scoring)
@@ -27,13 +22,14 @@ Researches real market demand, scores ideas on monetization and open-source impa
 - [Persistence](#persistence)
 - [Architecture](#architecture)
 - [CLI](#cli)
-- [Requirements](#requirements)
 - [Contributing](#contributing)
+- [Acknowledgements](#acknowledgements)
 - [License](#license)
 
 ## The Problem
 
 You have Claude Code tokens but no idea what to build. Existing idea generators are either:
+
 - **Static lists** frozen in time (91K+ stars on app-ideas, never updated)
 - **Thin AI wrappers** that hallucinate untested concepts
 - **Validation-only tools** that require you already have an idea
@@ -71,21 +67,95 @@ IdeaBox researches **real demand signals** from 6 source categories — HN, GitH
 claude plugin add pawanpaudel93/ideabox
 ```
 
-### Via npm
+### Via npm (recommended for CLI install)
 
 ```bash
 npx ideabox init
 ```
 
+This copies skills into your project's `.claude/skills/` directory with all phase files and references bundled into each skill.
+
 ### From Source
 
 ```bash
 git clone https://github.com/pawanpaudel93/ideabox.git
 cd ideabox
-npm install
+pnpm install
 node bin/cli.mjs init
 ```
 
+## Requirements
+
+- [Claude Code](https://code.claude.com) or [Codex CLI](https://github.com/openai/codex)
+- Node.js >= 18
+- `gh` CLI (recommended — higher GitHub API rate limits)
+
+## Permissions
+
+IdeaBox uses these tools during its pipeline:
+
+| Tool | Phase | Purpose |
+|------|-------|---------|
+| `WebSearch` | Research | Search HN, GitHub, Reddit for demand signals |
+| `WebFetch` | Research | Hit structured APIs (HN Algolia, npm, Reddit JSON) |
+| `Agent` | Research, Build | Parallel subagents for research and implementation |
+| `Read` | All | Read files, state, research artifacts |
+| `Write` | All | Write phase outputs, state, profiles |
+| `Edit` | Build, Polish | Modify source code |
+| `Bash` | Build, QA, Ship | Run tests, git commands, CLI tools |
+| `Glob` / `Grep` | Build, QA | Search codebase |
+
+### Claude Code
+
+**Auto mode (recommended)** — press `Shift+Tab` to toggle. Approves most tool calls with background safety checks. Requires Team/Enterprise/API plan with Sonnet 4.6+ or Opus 4.6+.
+
+**Accept edits mode** — press `Shift+Tab` to select `acceptEdits`. Auto-approves file changes, prompts for Bash and network.
+
+**Custom permission rules** — add to `.claude/settings.json` (project) or `~/.claude/settings.json` (global):
+
+```json
+{
+  "permissions": {
+    "allow": [
+      "WebSearch",
+      "WebFetch",
+      "Glob",
+      "Grep",
+      "Read",
+      "Edit",
+      "Write",
+      "Bash(git *)",
+      "Bash(gh *)",
+      "Bash(npm *)",
+      "Bash(pnpm *)",
+      "Bash(node *)"
+    ]
+  }
+}
+```
+
+> **Note:** `deny` rules take precedence over `allow`. Protected paths (`.git/`, `.env`) always prompt regardless of mode.
+
+### Codex CLI
+
+```bash
+# Recommended: workspace-write sandbox with auto-approvals
+codex --full-auto "run ideabox research"
+
+# More control: approval mode + sandbox separately
+codex -a untrusted -s workspace-write "run ideabox"
+
+# Maximum access (use with caution)
+codex --yolo "run ideabox"
+```
+
+| Codex Flag | Claude Code Equivalent |
+|-----------|----------------------|
+| `-a on-request` | `default` mode |
+| `-a untrusted` | `acceptEdits` mode |
+| `--full-auto` | `auto` mode (sandboxed) |
+| `--yolo` | `bypassPermissions` mode |
+
 ## Usage
 
 ```
@@ -93,8 +163,7 @@ node bin/cli.mjs init
 /ideabox research               # Browse ideas without building
 /ideabox backlog                # View saved ideas
 /ideabox backlog compare 1 3    # Head-to-head idea comparison
-/ideabox profile                # Set up or edit your profile
-/ideabox profile                # Reset learning data (choose "reset learning")
+/ideabox profile                # Set up, edit, or reset learning
 ```
 
 ### First Run
@@ -226,15 +295,19 @@ All research uses **token-efficient structured APIs** (not generic web scraping)
 IdeaBox gets smarter the more you use it through **4 auto-iterating loops**:
 
 ### 1. Preference Learning (immediate)
+
 Tracks which ideas you accept, dismiss, start, complete, or abandon. Updates category/complexity/monetization scores. Anti-echo-chamber: always reserves 10% for diverse suggestions.
 
 ### 2. Source Quality Tracking (after 5 sessions)
+
 Tracks which research sources contribute to ideas you actually choose. High-performing sources get expanded queries; low-performing sources get reduced. Scores clamped to [0.1, 1.0] — nothing fully eliminated.
 
 ### 3. Scoring Weight Adaptation (after 10 outcomes)
+
 Tracks which scoring dimensions predict ideas that actually ship. Auto-adjusts weights (e.g., demand signal matters more than feasibility for you). Clamped to [0.5x, 1.5x] to prevent over-fitting.
 
 ### 4. Query Evolution (after 3 uses per query)
+
 Tracks which search queries return useful results. Productive queries spawn variations; dead queries retire. Retired queries archived for trend analysis.
 
 **Reset:** `/ideabox profile` -> choose "reset learning"
@@ -269,15 +342,15 @@ ideabox/
 │   ├── ideabox/                 # Main skill (/ideabox)
 │   │   ├── SKILL.md             # Pipeline router (~3KB)
 │   │   ├── phases/
-│   │   │   ├── 01-research.md   # Parallel API research
-│   │   │   ├── 02-brainstorm.md # Design spec creation
-│   │   │   ├── 03-plan.md       # TDD implementation plan
-│   │   │   ├── 04-build.md      # Subagent-driven development
-│   │   │   ├── 05-qa.md         # Health score testing
-│   │   │   ├── 06-polish.md     # Visual QA + AI slop detection
-│   │   │   ├── 07-ship.md       # PR + version + changelog
-│   │   │   ├── 08-post-ship.md  # Canary + docs sync
-│   │   │   └── 09-learn.md      # Self-improvement
+│   │   │   ├── 01-research.md
+│   │   │   ├── 02-brainstorm.md
+│   │   │   ├── 03-plan.md
+│   │   │   ├── 04-build.md
+│   │   │   ├── 05-qa.md
+│   │   │   ├── 06-polish.md
+│   │   │   ├── 07-ship.md
+│   │   │   ├── 08-post-ship.md
+│   │   │   └── 09-learn.md
 │   │   └── references/
 │   │       ├── research-sources.md
 │   │       ├── scoring-rubric.md
@@ -302,86 +375,13 @@ npx ideabox doctor      # Run health checks
 ```
 
 The `doctor` command verifies:
+
 - Node.js >= 18
 - Data directory (`~/.ideabox/`)
 - Profile configured
 - Skills installed (checks Claude Code and Codex paths)
 - CLI availability
 
-## Requirements
-
-- [Claude Code](https://code.claude.com) or [Codex CLI](https://github.com/openai/codex)
-- Node.js >= 18
-- `gh` CLI recommended (for GitHub API access with higher rate limits)
-
-## Permissions
-
-IdeaBox uses these tools during its pipeline:
-
-| Tool | Phase | Purpose |
-|------|-------|---------|
-| `WebSearch` | Research | Search HN, GitHub, Reddit for demand signals |
-| `WebFetch` | Research | Hit structured APIs (HN Algolia, npm, Reddit JSON) |
-| `Agent` | Research, Build | Parallel subagents for research and implementation |
-| `Read` | All | Read files, state, research artifacts |
-| `Write` | All | Write phase outputs, state, profiles |
-| `Edit` | Build, Polish | Modify source code |
-| `Bash` | Build, QA, Ship | Run tests, git commands, CLI tools |
-| `Glob` / `Grep` | Build, QA | Search codebase |
-
-### Claude Code
-
-**Option 1: Auto mode (recommended)** — press `Shift+Tab` to toggle. Approves most tool calls with background safety checks. Requires Team/Enterprise/API plan with Sonnet 4.6+ or Opus 4.6+.
-
-**Option 2: Accept edits mode** — press `Shift+Tab` to select `acceptEdits`. Auto-approves file changes, prompts for Bash and network.
-
-**Option 3: Custom permission rules** — add to `.claude/settings.json` (project) or `~/.claude/settings.json` (global):
-
-```json
-{
-  "permissions": {
-    "allow": [
-      "WebSearch",
-      "WebFetch",
-      "Glob",
-      "Grep",
-      "Read",
-      "Edit",
-      "Write",
-      "Bash(git *)",
-      "Bash(gh *)",
-      "Bash(npm *)",
-      "Bash(pnpm *)",
-      "Bash(node *)"
-    ]
-  }
-}
-```
-
-> **Note:** `deny` rules take precedence over `allow`. Protected paths (`.git/`, `.env`) always prompt regardless of mode.
-
-### Codex CLI
-
-Codex CLI uses different permission flags:
-
-```bash
-# Recommended: workspace-write sandbox with auto-approvals
-codex --full-auto "run ideabox research"
-
-# More control: approval mode + sandbox separately
-codex -a untrusted -s workspace-write "run ideabox"
-
-# Maximum access (use with caution)
-codex --yolo "run ideabox"
-```
-
-| Codex Flag | Claude Code Equivalent |
-|-----------|----------------------|
-| `-a on-request` | `default` mode |
-| `-a untrusted` | `acceptEdits` mode |
-| `--full-auto` | `auto` mode (sandboxed) |
-| `--yolo` | `bypassPermissions` mode |
-
 ## Contributing
 
 Contributions welcome! Please:
@@ -397,7 +397,7 @@ Contributions welcome! Please:
 IdeaBox's pipeline design was inspired by patterns and approaches from these projects:
 
 | Project | Inspiration |
-|---------|--------------|
+|---------|-------------|
 | [superpowers](https://github.com/obra/superpowers) | Brainstorming workflow, TDD discipline, writing-plans, subagent-driven development, verification-before-completion, systematic debugging |
 | [gstack](https://github.com/garrytan/gstack) | QA health score rubric, design review with AI slop detection, ship workflow, canary monitoring, office-hours forcing questions |
 | [oh-my-claudecode](https://github.com/Yeachan-Heo/oh-my-claudecode) | Deep-interview validation, artifact-based phase skipping, autopilot phase chaining |

package/bin/cli.mjs

@@ -18,11 +18,58 @@ const __dirname = dirname(__filename);
 const ROOT = resolve(__dirname, "..");
 
 const SKILLS = ["ideabox", "research", "backlog", "profile"];
+const VALID_TARGETS = ["claude", "codex", "codex-project"];
 const MARKER_START = "<!-- ideabox:start -->";
 const MARKER_END = "<!-- ideabox:end -->";
+const INSTALL_MARKER = ".ideabox-installed";
 const HOME = homedir();
 const DATA_DIR = join(HOME, ".ideabox");
 
+// ASCII-safe status indicators (work on all terminals)
+const OK = "[OK]";
+const FAIL = "[X] ";
+const WARN = "[!] ";
+const INFO = "[-] ";
+
+function readPackageVersion() {
+  try {
+    const pkg = JSON.parse(readFileSync(join(ROOT, "package.json"), "utf-8"));
+    return pkg.version || "unknown";
+  } catch {
+    return "unknown";
+  }
+}
+
+function printHelp() {
+  console.log(`
+ideabox v${readPackageVersion()}
+Data-driven project idea engine for coding agents
+
+Usage:
+  ideabox <command> [targets...] [options]
+
+Commands:
+  init [targets...]       Install skills to current project
+  uninstall [targets...]  Remove skills from current project
+  doctor                  Run health checks
+  --version, -v           Show version
+  --help, -h              Show this help
+
+Targets:
+  claude          Install to .claude/skills/ (project)
+  codex           Install to ~/.codex/skills/ (user)
+  codex-project   Install to .codex/skills/ (project)
+
+If no target is specified, auto-detects based on installed CLIs.
+
+Examples:
+  ideabox init              Auto-detect and install
+  ideabox init claude       Install only to Claude Code
+  ideabox init claude codex Install to both Claude and Codex
+  ideabox doctor            Check installation health
+`);
+}
+
 function hasCommand(cmd) {
   try {
     const check = process.platform === "win32" ? `where ${cmd}` : `command -v ${cmd}`;
@@ -44,6 +91,15 @@ function autoDetect() {
   return targets;
 }
 
+function validateTargets(targets) {
+  const invalid = targets.filter((t) => !VALID_TARGETS.includes(t));
+  if (invalid.length > 0) {
+    console.error(`\nError: Unknown target(s): ${invalid.join(", ")}`);
+    console.error(`Valid targets: ${VALID_TARGETS.join(", ")}\n`);
+    process.exit(1);
+  }
+}
+
 function getSkillsDir(target) {
   switch (target) {
     case "claude":
@@ -53,11 +109,12 @@ function getSkillsDir(target) {
     case "codex-project":
       return join(process.cwd(), ".codex", "skills");
     default:
-      return join(process.cwd(), ".claude", "skills");
+      throw new Error(`Unknown target: ${target}`);
   }
 }
 
 function getInstructionFile(target) {
+  // Codex reads AGENTS.md from project root even for global install
   if (target === "codex" || target === "codex-project") {
     return join(process.cwd(), "AGENTS.md");
   }
@@ -132,6 +189,9 @@ function installSkills(target) {
       }
     }
 
+    // Write marker file so uninstall knows this skill was ours
+    writeFileSync(join(dest, INSTALL_MARKER), `ideabox v${readPackageVersion()}\n`);
+
     console.log(`  Installed ${skill}`);
   }
 }
@@ -151,7 +211,7 @@ function install(targets) {
     console.log(`Target: ${target}`);
     installSkills(target);
     mergeInstructionFile(target);
-    console.log(`  Merged instruction file for ${target}`);
+    console.log(`  Merged instruction file: ${getInstructionFile(target)}`);
   }
 
   console.log("\nDone! Run /ideabox to get started.\n");
@@ -165,9 +225,14 @@ function uninstall(targets) {
 
     for (const skill of SKILLS) {
       const dest = join(skillsDir, skill);
-      if (existsSync(dest)) {
+      if (!existsSync(dest)) continue;
+
+      // Only remove skills that have our install marker
+      if (existsSync(join(dest, INSTALL_MARKER))) {
         rmSync(dest, { recursive: true });
         console.log(`  Removed ${skill}`);
+      } else {
+        console.log(`  Skipped ${skill} (not installed by ideabox)`);
       }
     }
 
@@ -184,58 +249,69 @@ function uninstall(targets) {
   console.log(`\nNote: ${DATA_DIR} was kept. Delete it manually if you want to remove all data.\n`);
 }
 
+function checkSkillsAt(label, skillsDir) {
+  if (!existsSync(skillsDir)) return null;
+  const installed = SKILLS.filter((s) => existsSync(join(skillsDir, s)));
+  return { label, path: skillsDir, installed };
+}
+
 function doctor() {
   console.log("\nIdeaBox -- Health Check\n");
   let ok = true;
 
   // Check 1: Node version
-  const nodeVersion = parseInt(process.version.slice(1));
+  const nodeVersion = parseInt(process.version.slice(1), 10);
   if (nodeVersion >= 18) {
-    console.log(`  ✓ Node.js ${process.version}`);
+    console.log(`  ${OK} Node.js ${process.version}`);
   } else {
-    console.log(`  ✗ Node.js ${process.version} (need >= 18)`);
+    console.log(`  ${FAIL} Node.js ${process.version} (need >= 18)`);
     ok = false;
   }
 
   // Check 2: Data directory
   if (existsSync(DATA_DIR)) {
-    console.log(`  ✓ Data directory exists (${DATA_DIR})`);
+    console.log(`  ${OK} Data directory exists (${DATA_DIR})`);
   } else {
-    console.log(`  ✗ Data directory missing (${DATA_DIR})`);
+    console.log(`  ${FAIL} Data directory missing (${DATA_DIR})`);
     ok = false;
   }
 
   // Check 3: Profile
   const profilePath = join(DATA_DIR, "profile.json");
   if (existsSync(profilePath)) {
-    console.log(`  ✓ Profile configured`);
+    console.log(`  ${OK} Profile configured`);
   } else {
-    console.log(`  ○ No profile yet (will be created on first /ideas run)`);
+    console.log(`  ${INFO} No profile yet (will be created on first /ideabox run)`);
   }
 
-  // Check 4: Skills installed (check both Claude and Codex paths)
-  const claudeSkills = join(process.cwd(), ".claude", "skills");
-  const codexSkills = join(HOME, ".codex", "skills");
-  const claudeInstalled = SKILLS.filter((s) => existsSync(join(claudeSkills, s)));
-  const codexInstalled = SKILLS.filter((s) => existsSync(join(codexSkills, s)));
-  const installed = claudeInstalled.length >= codexInstalled.length ? claudeInstalled : codexInstalled;
-  const installTarget = claudeInstalled.length >= codexInstalled.length ? "Claude" : "Codex";
-  if (installed.length === SKILLS.length) {
-    console.log(`  ✓ All ${SKILLS.length} skills installed (${installTarget})`);
-  } else if (installed.length > 0) {
-    console.log(`  △ ${installed.length}/${SKILLS.length} skills installed (${installed.join(", ")})`);
-  } else {
-    console.log(`  ✗ No skills installed in current project`);
+  // Check 4: Skills installed (check all target paths)
+  const paths = [
+    checkSkillsAt("Claude (.claude/skills)", join(process.cwd(), ".claude", "skills")),
+    checkSkillsAt("Codex user (~/.codex/skills)", join(HOME, ".codex", "skills")),
+    checkSkillsAt("Codex project (./.codex/skills)", join(process.cwd(), ".codex", "skills")),
+  ].filter(Boolean);
+
+  const anyInstalled = paths.some((p) => p.installed.length > 0);
+  if (!anyInstalled) {
+    console.log(`  ${FAIL} No skills installed in current project`);
     ok = false;
+  } else {
+    for (const p of paths) {
+      if (p.installed.length === SKILLS.length) {
+        console.log(`  ${OK} ${p.label}: ${p.installed.length}/${SKILLS.length} skills`);
+      } else if (p.installed.length > 0) {
+        console.log(`  ${WARN} ${p.label}: ${p.installed.length}/${SKILLS.length} skills (${p.installed.join(", ")})`);
+      }
+    }
   }
 
   // Check 5: Claude Code / Codex available
   if (hasCommand("claude")) {
-    console.log(`  ✓ Claude Code CLI detected`);
+    console.log(`  ${OK} Claude Code CLI detected`);
   } else if (hasCommand("codex")) {
-    console.log(`  ✓ Codex CLI detected`);
+    console.log(`  ${OK} Codex CLI detected`);
   } else {
-    console.log(`  ○ No supported CLI found (optional)`);
+    console.log(`  ${INFO} No supported CLI found (optional)`);
   }
 
   console.log(ok ? "\n  All checks passed!\n" : "\n  Some issues found.\n");
@@ -244,13 +320,26 @@ function doctor() {
 // --- Main ---
 function parseTargets(args) {
   const explicit = args.slice(1).filter((a) => !a.startsWith("-") && a !== "--");
-  return explicit.length > 0 ? explicit : autoDetect();
+  if (explicit.length > 0) {
+    validateTargets(explicit);
+    return explicit;
+  }
+  return autoDetect();
 }
 
 const args = process.argv.slice(2);
 const command = args[0] || "init";
 
 switch (command) {
+  case "--version":
+  case "-v":
+    console.log(`ideabox v${readPackageVersion()}`);
+    break;
+  case "--help":
+  case "-h":
+  case "help":
+    printHelp();
+    break;
   case "init":
     install(parseTargets(args));
     break;
@@ -261,7 +350,7 @@ switch (command) {
     doctor();
     break;
   default:
-    console.log(`Unknown command: ${command}`);
-    console.log("Usage: ideabox-agent [init|uninstall|doctor]");
+    console.log(`Unknown command: ${command}\n`);
+    printHelp();
     process.exit(1);
 }

package/package.json

@@ -1,12 +1,13 @@
 {
   "name": "ideabox",
-  "version": "1.0.0",
+  "version": "1.0.2",
   "description": "Data-driven project idea engine for coding agents. Research, brainstorm, plan, and build — all from one command.",
   "type": "module",
   "bin": {
     "ideabox": "bin/cli.mjs"
   },
   "files": [
+    ".claude-plugin/",
     "bin/",
     "skills/",
     "CLAUDE.md",
@@ -33,7 +34,7 @@
     "bumpp": "^10.4.1"
   },
   "scripts": {
-    "version": "node --input-type=commonjs -e \"const fs=require('fs');const p=JSON.parse(fs.readFileSync('.claude-plugin/plugin.json'));p.version=require('./package.json').version;fs.writeFileSync('.claude-plugin/plugin.json',JSON.stringify(p,null,2)+'\\n');\"",
-    "release": "bumpp && pnpm publish"
+    "version": "node --input-type=commonjs -e \"const fs=require('fs');const p=JSON.parse(fs.readFileSync('.claude-plugin/plugin.json'));p.version=require('./package.json').version;fs.writeFileSync('.claude-plugin/plugin.json',JSON.stringify(p,null,2)+'\\n');\" && git add .claude-plugin/plugin.json",
+    "release": "bumpp --all && pnpm publish"
   }
 }

package/skills/backlog/SKILL.md

@@ -19,15 +19,15 @@ Read `~/.ideabox/ideas.jsonl`. Each line is a JSON object. There are two record
 To get the current status of an idea: find the latest `status_update` for that `idea_id`. If no status_update exists, use the idea record's `status` field.
 
 Parse arguments from the user's command:
-- `/ideas backlog` — show all ideas grouped by status
-- `/ideas backlog saved` — show only saved ideas
-- `/ideas backlog dismissed` — show only dismissed ideas
-- `/ideas backlog built` — show only built ideas
+- `/ideabox backlog` — show all ideas grouped by status
+- `/ideabox backlog saved` — show only saved ideas
+- `/ideabox backlog dismissed` — show only dismissed ideas
+- `/ideabox backlog built` — show only built ideas
 
 ## Display Format
 
 ### If ideas.jsonl does not exist or is empty
-"No ideas in your backlog yet. Run `/ideas` to research your first batch!"
+"No ideas in your backlog yet. Run `/ideabox` to research your first batch!"
 
 ### If ideas exist
 
@@ -50,14 +50,14 @@ Group by current status and display:
 |---|------|-------|---------|
 
 ### Dismissed
-Showing {count} dismissed ideas. Use `/ideas backlog dismissed` for details.
+Showing {count} dismissed ideas. Use `/ideabox backlog dismissed` for details.
 ```
 
 ## Actions
 
 After displaying, offer:
 - "Pick a number to brainstorm and build that idea"
-- "Or run `/ideas` for fresh research"
+- "Or run `/ideabox` for fresh research"
 
 If the user picks a saved idea by number:
 1. Read the full idea record from `ideas.jsonl`
@@ -69,23 +69,23 @@ If the user picks a saved idea by number:
       {"type":"status_update","idea_id":"{id}","old_status":"saved","new_status":"planned","timestamp":"{ISO}"}
       ```
    b. Create `.ideabox/` and `.ideabox/session/` directories if they don't exist
-   c. Initialize `.ideabox/state.json`:
+   c. Write the idea details to `.ideabox/session/01-research.md` first (so phase 02 has context and state references a real file)
+   d. Initialize `.ideabox/state.json` with the artifact path populated:
       ```json
       {
         "session_id": "sess_YYYYMMDD_HHMMSS",
         "current_phase": "02-brainstorm",
-        "phases_completed": [],
+        "phases_completed": ["01-research"],
         "idea": {"id": "{id}", "title": "{title}", "problem": "{problem}"},
-        "artifacts": {},
+        "artifacts": {"research": ".ideabox/session/01-research.md"},
         "started_at": "{ISO}"
       }
       ```
-   d. Write the idea details to `.ideabox/session/01-research.md` (so phase 02 has context)
    e. Read `${CLAUDE_SKILL_DIR}/phases/02-brainstorm.md` and follow it
 
 ## Comparison Mode
 
-If the user says `/ideas backlog compare N M` (e.g., `compare 1 3`):
+If the user says `/ideabox backlog compare N M` (e.g., `compare 1 3`):
 
 1. Load both idea records from `ideas.jsonl`
 2. Present using the visual comparison format from `${CLAUDE_SKILL_DIR}/references/scoring-rubric.md` (Head-to-Head section) — with score bars, evidence, and a recommendation

package/skills/ideabox/SKILL.md

@@ -10,14 +10,14 @@ argument-hint: "[research|backlog|profile]"
 
 # IdeaBox — Full Pipeline
 
-Research -> Present -> Pick -> Brainstorm -> Plan -> Build -> QA -> Polish -> Ship -> Post-Ship -> Learn
+Research -> Brainstorm -> Plan -> Build -> QA -> Polish -> Ship -> Post-Ship -> Learn
 
 ## Subcommand Routing
 
 Check if the user passed a subcommand:
-- `/ideabox research` -> invoke `ideabox:research` skill
-- `/ideabox backlog` -> invoke `ideabox:backlog` skill
-- `/ideabox profile` -> invoke `ideabox:profile` skill
+- `/ideabox research` -> invoke the `research` skill (namespaced as `ideabox:research` when installed as plugin)
+- `/ideabox backlog` -> invoke the `backlog` skill (namespaced as `ideabox:backlog` when installed as plugin)
+- `/ideabox profile` -> invoke the `profile` skill (namespaced as `ideabox:profile` when installed as plugin)
 - `/ideabox` (no subcommand) -> continue with pipeline below
 - `/ideabox [anything else]` -> respond: "Unknown subcommand. Available: `/ideabox`, `/ideabox research`, `/ideabox backlog`, `/ideabox profile`"
 
@@ -33,10 +33,13 @@ Ask: "You have an in-progress session (Phase: {current_phase}, Idea: {idea.title
 **If no state.json or starting fresh:**
 1. Create `.ideabox/` and `.ideabox/session/` directories
 2. Add `.ideabox/` to the project's `.gitignore` if not already present (session state should not be committed)
-3. Check for existing artifacts (phase skipping):
-   - If `~/.ideabox/research/` has a file from today -> ask: "You already researched today. Use those results or research fresh?"
-   - If `.ideabox/session/02-brainstorm-spec.md` exists -> ask: "Found an existing spec. Resume from planning?"
-   - If `.ideabox/session/03-plan.md` exists -> ask: "Found an existing plan. Resume from building?"
+3. Check for existing artifacts (phase skipping). Scan `.ideabox/session/` for any of these files and offer to skip ahead:
+   - `01-research.md` exists -> ask: "Found research results. Resume from brainstorm?"
+   - `02-brainstorm-spec.md` exists -> ask: "Found an existing spec. Resume from planning?"
+   - `03-plan.md` exists -> ask: "Found an existing plan. Resume from building?"
+   - `04-handoff.md` exists -> ask: "Found build results. Resume from QA?"
+   - `05-qa-report.md` exists -> ask: "Found QA results. Resume from polish?"
+   - Also check `~/.ideabox/research/` for a file from today -> ask: "You already researched today. Use those results or research fresh?"
 4. Initialize `.ideabox/state.json` with the appropriate starting phase:
 ```json
 {
@@ -87,7 +90,7 @@ After each phase completes and its gate condition is met:
    - **yes**: read next phase file and follow it
    - **revise**: stay in current phase, user explains what to change
    - **skip**: jump to phase 09 (learn) to save progress, then end
-   - **abort**: save state, end session (can resume later with `/ideas`)
+   - **abort**: save state, end session (can resume later with `/ideabox`)
 
 ## Phase Rules
 
@@ -100,11 +103,11 @@ After each phase completes and its gate condition is met:
 - **Re-read state** before every phase transition to prevent drift
 - **Each phase writes output** to `.ideabox/session/` (e.g., `01-research.md`, `02-brainstorm-spec.md`, `03-plan.md`, `05-qa-report.md`)
 - **Next phase reads** only its own instructions + the previous phase's output (not all phases)
-- **If context is running low:** save state to `.ideabox/state.json`, tell user: "Context getting full. Run `/ideas` in a new session to resume from phase {current_phase}."
+- **If context is running low:** save state to `.ideabox/state.json`, tell user: "Context getting full. Run `/ideabox` in a new session to resume from phase {current_phase}."
 
 ## Error Handling
 
 - **Phase gate fails:** stay in current phase, explain what's needed to pass
-- **User aborts:** save state, they resume later with `/ideas`
+- **User aborts:** save state, they resume later with `/ideabox`
 - **Subagent fails:** note the failure, offer to retry or skip
 - **No profile:** run profile setup inline (don't tell user to run another command)

package/skills/ideabox/phases/01-research.md

@@ -11,9 +11,9 @@ Research project ideas from multiple data sources using parallel subagents. Pres
 
 Read `${CLAUDE_SKILL_DIR}/references/self-improvement.md` for the full self-improvement engine specification, then load adaptive data:
 
-1. **Source quality** — read `~/.ideabox/source-quality.jsonl` (if exists, 5+ entries). Compute source scores. Allocate more queries to high-quality sources, reduce queries for low-quality sources.
-2. **Scoring weights** — read `~/.ideabox/scoring-feedback.jsonl` (if exists, 10+ outcomes). Compute adapted dimension weights. If insufficient data, use default equal weights (1.0 each).
-3. **Query evolution** — read `~/.ideabox/query-performance.jsonl` (if exists). Build the active query set: drop RETIRED queries, add variations for PRODUCTIVE queries. If no data, use default queries from `${CLAUDE_SKILL_DIR}/references/research-sources.md`.
+1. **Source quality** — read `~/.ideabox/source-quality.jsonl` if it exists. Compute source scores only if the file has 5+ entries; otherwise use default equal allocation. Allocate more queries to high-quality sources, reduce queries for low-quality sources.
+2. **Scoring weights** — read `~/.ideabox/scoring-feedback.jsonl` if it exists. Compute adapted dimension weights only if the file has 10+ outcomes; otherwise use default equal weights (1.0 each).
+3. **Query evolution** — read `~/.ideabox/query-performance.jsonl` if it exists. Build the active query set: drop RETIRED queries, add variations for PRODUCTIVE queries. If no data, use default queries from `${CLAUDE_SKILL_DIR}/references/research-sources.md`.
 
 If any self-improvement file doesn't exist or has insufficient data, use defaults silently — no error messages for new users.
 
@@ -60,7 +60,7 @@ Return JSON array with: source_category: 'indie_hacker', source_url, signal_type
 Return JSON array with: source_category: 'packages', source_url, signal_type, title, description, evidence, demand_score."
 
 **Subagent 6 — User's GitHub Profile:**
-"Fetch the GitHub profile for {profile.github_username}. Use WebFetch on `https://api.github.com/users/{username}/repos?sort=updated&per_page=30`. Extract primary languages, frameworks, project types, and notable repos with star counts.
+"Fetch the GitHub profile for {profile.github_username}. Prefer `gh api '/users/{username}/repos?sort=updated&per_page=30'` if `gh` CLI is available (5000 req/hr authenticated). Otherwise fall back to WebFetch on `https://api.github.com/users/{username}/repos?sort=updated&per_page=30` (60 req/hr unauthenticated). Extract primary languages, frameworks, project types, and notable repos with star counts.
 Return a JSON object: {languages: [...], frameworks: [...], project_types: [...], notable_repos: [{name, stars, description}]}"
 
 ## Step 2: Score & Filter
@@ -69,7 +69,7 @@ Once all subagents return:
 
 1. **Merge** all findings into a single list
 2. **Synthesize** — look for ideas that combine signals from multiple findings (same gap from different sources = one stronger idea, not two weak ones)
-3. **Cross-reference bonus** — ideas appearing across 2+ sources get +2 demand signal; 3+ sources get +3 (capped at 10)
+3. **Cross-reference bonus** — When rating demand signal, first rate based on the strongest single source (1-10), THEN add bonus: +2 if appeared in 2 sources, +3 if 3+ sources (capped at 10). Do not pre-inflate the base rating for multi-source evidence.
 4. **Apply agentic AI bonus** — ideas in agentic AI / MCP / AI tooling space get +2 trend momentum (capped at 10)
 5. **Filter history** — load dismissed/built ideas from `~/.ideabox/ideas.jsonl`, remove ideas with >70% title+problem similarity
 6. **Filter avoid topics** — remove ideas matching `profile.avoid_topics`
@@ -79,7 +79,7 @@ Once all subagents return:
    - Apply adapted scoring weights from Step 0 (if available). Multiply each dimension score by its weight before summing.
    - If using adapted weights, show them: "Scoring with adapted weights: Revenue 1.2x, Demand 1.3x, ..."
    - Compute weighted total
-8. **Filter** — remove ideas scoring below 25 (adjusted for weight changes)
+8. **Filter** — remove ideas scoring below 25 (scale is always 0-60 regardless of weight adaptation)
 9. **Rank** — sort by total score descending
 10. **Update profile stacks** — if GitHub subagent returned language data, update `~/.ideabox/profile.json` stacks field
 11. **Record query performance** — for each subagent's queries, append to `~/.ideabox/query-performance.jsonl`: query text, results count, useful results count, contributed to any presented idea
@@ -137,13 +137,13 @@ Then ask:
 ## Step 7: Handle User Choice
 
 **If user picks an idea to build:**
-1. Append the full idea record to `~/.ideabox/ideas.jsonl` with status "planned"
-2. Update `.ideabox/state.json`:
+1. Write research summary to `.ideabox/session/01-research.md` (idea details + evidence + scores)
+2. Append the full idea record to `~/.ideabox/ideas.jsonl` with status "planned"
+3. Update `.ideabox/state.json` (AFTER writing the artifact so state always points to a real file):
    - Set `idea` field with id, title, problem
    - Set `current_phase` to "02-brainstorm"
    - Add "01-research" to `phases_completed`
    - Set `artifacts.research` to `.ideabox/session/01-research.md`
-3. Write research summary to `.ideabox/session/01-research.md` (idea details + evidence + scores)
 4. Do NOT log to sessions.jsonl here — Phase 09 (Learn) handles the final session log with complete outcome data
 5. Proceed to Phase 02
 

package/skills/ideabox/phases/05-qa.md

@@ -78,6 +78,25 @@ Compute a health score (0-100) based on findings:
 - Medium: -8
 - Low: -3
 
+**Computation (in order):**
+1. Compute each category's score (0-100) based on issue count
+2. Compute `base_score = sum(category_score * weight)` across all 8 categories — yields 0-100
+3. For each distinct bug found, subtract its severity deduction from `base_score`
+4. Clamp final score to `[0, 100]` (never go negative)
+
+**Worked example:**
+- Console: 0 errors → 100 × 15% = 15.0
+- Links: 0 broken → 100 × 10% = 10.0
+- Visual: 2 issues → 70 × 10% = 7.0
+- Functional: 0 bugs → 100 × 20% = 20.0
+- UX: 1 issue → 70 × 15% = 10.5
+- Performance: 0 issues → 100 × 10% = 10.0
+- Content: 0 issues → 100 × 5% = 5.0
+- Accessibility: 3 issues → 70 × 15% = 10.5
+- `base_score` = 88.0
+- Found 1 critical, 2 high = -25 + -15 + -15 = -55
+- `final_score` = max(0, 88 - 55) = **33/100** (fails QA gate of 70)
+
 ## Step 4: Fix-Verify Loop
 
 For each bug found (prioritized by severity — critical first):

package/skills/ideabox/phases/07-ship.md

@@ -10,10 +10,12 @@ Create PR, bump version, update changelog, and prepare for deployment.
 
 ## Step 2: Merge Base Branch
 
-Ensure you're up to date with the base branch:
+Detect the base branch dynamically (don't assume `main`):
 ```bash
-git fetch origin main && git merge origin/main --no-edit
+BASE=$(git symbolic-ref refs/remotes/origin/HEAD 2>/dev/null | sed 's|refs/remotes/origin/||') || BASE="main"
+git fetch origin "$BASE" && git merge "origin/$BASE" --no-edit
 ```
+Common base branch names: `main`, `master`, `develop`. If detection fails or yields the wrong branch, ask the user.
 If merge conflicts: resolve them, run tests, commit.
 
 ## Step 3: Run Tests

package/skills/ideabox/phases/09-learn.md

@@ -4,16 +4,16 @@ Track what happened, update preferences, and improve future suggestions. This ph
 
 ## Step 1: Record Session Outcome
 
-Determine the session outcome based on which phases completed:
+Determine the session outcome based on which phases completed. **Read top-to-bottom and use the first row that matches** (the table is ordered most-specific first):
 
 | Phases Completed | Outcome | Event |
 |-----------------|---------|-------|
-| 01 only | browsed | User looked at ideas but didn't pick one |
-| 01-02 | brainstormed | User refined an idea but didn't plan/build |
-| 01-03 | planned | User planned but didn't build |
-| 01-04+ | started | User began building |
-| 01-07+ | shipped | User shipped the project |
 | 01-08+ | completed | Full cycle including post-ship |
+| 01-07+ | shipped | User shipped the project |
+| 01-04 to 01-06 | started | User began building but didn't ship |
+| 01-03 | planned | User planned but didn't build |
+| 01-02 | brainstormed | User refined an idea but didn't plan/build |
+| 01 only | browsed | User looked at ideas but didn't pick one |
 
 ## Step 2: Update Idea Status
 
@@ -165,7 +165,7 @@ Present to the user:
 - Exploration rate: {old_rate}% -> {new_rate}%
 - Total ideas seen: {total_interactions}
 
-Run `/ideas` again for your next project!
+Run `/ideabox` again for your next project!
 ```
 
 ## Step 8: Log Session

package/skills/ideabox/references/research-sources.md

@@ -6,7 +6,7 @@ description: Source catalog for research subagents. 6 parallel subagents launche
 
 # Research Sources
 
-This catalog defines all available research categories. During execution, **6 parallel subagents** are launched (categories 1-4 + 5 + 6). Categories 5, 7, 8, and 10 are consolidated into other subagents or used as filter steps, not standalone subagents.
+This catalog defines all available research categories. During execution, **6 parallel subagents** are launched mapping to categories 1 (Agentic AI), 2 (Pain Points), 3 (Trending), 4 (Indie Hacker), 6 (Packages), and 9 (GitHub Profile). Categories 5, 7, 8 are consolidated into other subagents. Category 10 (Dismissed filter) is applied as a post-processing step, not a subagent.
 
 ---
 

package/skills/ideabox/references/self-improvement.md

@@ -21,11 +21,15 @@ Append after each session:
 For each source, compute a quality score:
 
 ```
-contribution_rate = ideas_contributed_to_chosen / total_sessions_with_source
-outcome_weight = completed: 1.0, started: 0.6, planned: 0.3, dismissed: 0.0
-weighted_score = sum(contribution_rate * outcome_weight) / sessions_count
+For each session where this source was used:
+  per_session_score = (1 if contributed_to_chosen else 0) * outcome_weight
+  where outcome_weight = completed: 1.0, started: 0.6, planned: 0.3, dismissed: 0.0
 
-Final source_score = 0.3 * contribution_rate + 0.4 * weighted_score + 0.3 * avg_signals_per_session
+contribution_rate = sessions_contributed_to_chosen / total_sessions_with_source
+weighted_score = mean(per_session_score across all sessions for this source)
+avg_signals = mean(signals_found across all sessions for this source) / 10  # normalize to 0-1
+
+Final source_score = 0.3 * contribution_rate + 0.4 * weighted_score + 0.3 * avg_signals
 Clamp to [0.1, 1.0]
 ```
 
@@ -81,13 +85,20 @@ For each dimension:
   abandoned_avg = average score for ideas with outcome=abandoned/dismissed
   predictive_power = completed_avg - abandoned_avg
 
-Normalize predictive_power across dimensions so they sum to 6.0 (preserving max 60 total).
+Step 1: Scale predictive_power to initial weights so they sum to 6.0
+Step 2: Clamp each weight to [0.5, 1.5] to prevent over-fitting
+Step 3: Re-normalize the clamped weights so they sum to 6.0
+
+Final rule: total_score is always normalized to the 0-60 scale by computing
+  raw_weighted_sum = sum(dimension_score * weight) across 6 dimensions
+  normalized_total = raw_weighted_sum * 6 / sum(weights)
+This guarantees max score = 60 regardless of weight distribution.
 ```
 
 **Example adaptation:**
 - If `feasibility` strongly predicts completion (high for completed, low for abandoned): increase its weight
 - If `stack_fit` doesn't predict outcomes (similar scores for both): decrease its weight
-- Never let any dimension drop below 0.5 or exceed 1.5 (prevent over-fitting)
+- Weights clamped to [0.5, 1.5] — after clamping, re-normalized to sum to 6.0
 
 ### How Weights Apply
 
@@ -213,5 +224,5 @@ All three fall back to defaults when insufficient data exists. No degradation fo
 1. **Score clamping**: source scores [0.1, 1.0], dimension weights [0.5, 1.5]
 2. **Never eliminate**: low-scoring sources get reduced queries, not removed
 3. **Archive, don't delete**: retired queries kept in log for trend analysis
-4. **Periodic reset option**: user can run `/ideas profile` and choose "reset learning" to clear all feedback data and start fresh
+4. **Periodic reset option**: user can run `/ideabox profile` and choose "reset learning" to clear all feedback data and start fresh
 5. **Transparency**: every adapted weight and source score is shown to the user — no hidden adjustments

package/skills/profile/SKILL.md

@@ -4,7 +4,7 @@ description: >
   Manage your IdeaBox profile — interests, tech stacks, goals, and GitHub username.
   Use when asked to "set up ideabox", "edit ideabox profile", "change interests",
   "update my stack", or "ideabox profile".
-  This skill is also invoked automatically on first /ideas run if no profile exists.
+  This skill is also invoked automatically on first /ideabox run if no profile exists.
 ---
 
 # IdeaBox Profile Manager
@@ -94,4 +94,4 @@ The `stacks` array will be populated by the research phase when it scans the use
 
 The `category_scores`, `complexity_preference`, `monetization_preference`, `exploration_rate`, and `total_interactions` fields are used by the self-improving recommendation system (Phase 09 — Learn).
 
-After saving, confirm: "Profile saved! Run `/ideas` to get your first batch of project ideas."
+After saving, confirm: "Profile saved! Run `/ideabox` to get your first batch of project ideas."

package/skills/research/SKILL.md

@@ -44,11 +44,11 @@ Append to `~/.ideabox/sessions.jsonl`:
   "session_id": "sess_YYYYMMDD_HHMMSS",
   "timestamp": "{ISO}",
   "mode": "research",
-  "sources_searched": "{actual count of subagents that returned results}",
-  "ideas_generated": "{count}",
-  "ideas_presented": "{count}",
-  "ideas_saved": "{count}",
-  "ideas_dismissed": "{count}"
+  "sources_searched": N,
+  "ideas_generated": N,
+  "ideas_presented": N,
+  "ideas_saved": N,
+  "ideas_dismissed": N
 }
 ```