kodelyth-ecc 1.5.9 → 1.7.0

package/wiki/FAQ.md

@@ -6,7 +6,7 @@
 
 **What is Kodelyth ECC?**
 
-An AI coding toolkit that installs into Claude Code, Windsurf, Cursor, Codex CLI, Google Antigravity, and OpenCode. It adds 62 specialist agents, 188 skills, 80 slash commands, 20+ automation hooks, and a compound learning system.
+An AI coding toolkit (v1.7.0) that installs into 12 AI coding platforms. It adds 70 specialist agents (including 8 adversarial devil-mode agents), 194 skills, 97 slash commands, 20+ automation hooks, and a compound learning system. Zero cloud, zero telemetry, all local.
 
 **Is it free?**
 
@@ -18,6 +18,7 @@ No. Everything is local:
 - Memory lives in `~/.kodelyth/memory/memories.jsonl`
 - Lessons live in `tasks/lessons.md` in your project
 - Token logs live in `~/.claude/logs/token-usage.jsonl`
+- MCP server runs on localhost only
 
 Nothing is transmitted to any server.
 
@@ -25,6 +26,10 @@ Nothing is transmitted to any server.
 
 Node.js 18 or higher.
 
+**What platforms does ECC support?**
+
+12 platforms: Claude Code, Windsurf (project/home), Cursor (project), Google Antigravity, Codex CLI (home), OpenCode, Cline, Roocode, Aider, Kimi, Gemini (project/home).
+
 ---
 
 ## Memory System
@@ -45,6 +50,10 @@ The `auto-recall.js` UserPromptSubmit hook runs BM25 search on every message you
 
 Yes. Edit `~/.kodelyth/memory/memories.jsonl` directly — it is plain JSONL. Or use `/memory delete <id>`.
 
+**What is self-evolving memory?**
+
+Memory in v1.7.0 now evolves. Run `npx kodelyth-ecc evolve analyze` to scan memory for patterns, extract meta-insights, and surface high-impact solutions. Memories can be tagged by domain, and the system learns which solutions get reused most often.
+
 **What is the difference between memory and lessons?**
 
 | | Memory | Lessons |
@@ -54,6 +63,7 @@ Yes. Edit `~/.kodelyth/memory/memories.jsonl` directly — it is plain JSONL. Or
 | Format | JSONL, structured | Markdown, human-readable |
 | Content | Past solutions, patterns | Hard rules from your corrections |
 | Injected | Per-prompt (BM25 match) | Session start (always) |
+| Evolves | Yes (v1.7.0+) | Manual edit |
 
 ---
 
@@ -64,8 +74,8 @@ Yes. Edit `~/.kodelyth/memory/memories.jsonl` directly — it is plain JSONL. Or
 Three layers that make Claude increasingly match your style:
 
 1. **Project Lessons** — corrections you give Claude are encoded as hard rules in `tasks/lessons.md` and applied every session
-2. **Global Memory** — past solutions recalled cross-project
-3. **Semantic Routing** — 62 specialists routing from intent
+2. **Global Memory** — past solutions recalled cross-project, auto-evolving
+3. **Semantic Routing** — 70 specialists routing from intent
 
 **How do lessons get written?**
 
@@ -81,15 +91,172 @@ Yes. `tasks/lessons.md` is plain Markdown. Add, edit, or delete rules freely. Th
 
 ---
 
+## MCP Server (v1.7.0)
+
+**What is the Kodelyth ECC MCP server?**
+
+A Model Context Protocol server that exposes 16 tools, 6 prompts, and 370+ resources. Run `npx kodelyth-ecc mcp` to start the server locally on a configurable port.
+
+**What tools does it provide?**
+
+16 tools covering agent invocation, memory management, skill loading, command execution, swarm orchestration, cost routing, and session analysis.
+
+**What prompts are available?**
+
+6 system prompts for common workflows: intent routing setup, memory protocol, security analysis, code architecture, testing strategy, and incident response.
+
+**How do I configure the MCP server?**
+
+```bash
+npx kodelyth-ecc mcp --port 3000
+npx kodelyth-ecc mcp --config ~/.kodelyth/mcp.json
+```
+
+Can also set `MCP_PORT` environment variable.
+
+**Is the MCP server secure?**
+
+Yes. Listens on localhost only. No external network access. All operations are read-only from disk unless explicitly writing to `tasks/lessons.md` or memory files.
+
+---
+
+## Local Dashboard (v1.7.0)
+
+**What is the Kodelyth dashboard?**
+
+A local web-based dashboard running on localhost. No analytics, no external calls. Five tabs: Overview, Memory, Evolve, Catalog, Sessions.
+
+**How do I launch it?**
+
+```bash
+npx kodelyth-ecc dashboard
+```
+
+Opens automatically at `http://localhost:5173` (or your configured port).
+
+**What can I do on each tab?**
+
+- **Overview** — Agent counts, installed components, config status
+- **Memory** — View, search, tag, and delete stored memories
+- **Evolve** — Analyze memory for patterns, surface high-impact solutions
+- **Catalog** — Browse all 70 agents and 194 skills with descriptions
+- **Sessions** — Replay session transcripts, see token usage, review lessons captured
+
+---
+
+## Devil-Mode Agents (v1.7.0)
+
+**What is devil-mode?**
+
+Dedicated red-team mode that fires 4-8 adversarial agents to stress-test your code from an attacker's perspective. Find vulnerabilities before bad actors do.
+
+**How do I invoke it?**
+
+```bash
+/devil-mode                 # Top 4: injection, supply chain, secrets, license
+/devil-mode --all          # All 8: adds jailbreak, theft, backdoors, chaos
+```
+
+**What are the 8 agents?**
+
+1. `prompt-injection-hunter` — AI injection vectors
+2. `supply-chain-auditor` — Dependency and artifact vulnerabilities
+3. `secret-hunter` — Hardcoded credentials
+4. `license-violation-finder` — GPL and license issues
+5. `jailbreak-tester` — Safety bypass testing
+6. `code-stealer-detector` — Exfiltration vectors
+7. `backdoor-hunter` — Inserted vulnerabilities
+8. `chaos-engineer` — Failure mode analysis
+
+**Is devil-mode destructive?**
+
+No. It analyzes code and reports findings. It does not modify anything or access external systems. Results are advisory — you decide what to fix.
+
+---
+
+## Swarm Orchestrator (v1.7.0)
+
+**What is the swarm orchestrator?**
+
+A system for running multiple agents in parallel on complex tasks. Agents communicate their findings and coordinate work distribution.
+
+**How do I use it?**
+
+```bash
+npx kodelyth-ecc swarm --task "audit my codebase for security + performance issues"
+```
+
+Also available as the `/swarm-orchestrator` skill.
+
+**How is it different from parallel commands?**
+
+Parallel commands fire fixed teams (e.g., `/team-review` always fires the same 4 agents). Swarm takes a free-form task, analyzes it, and dynamically selects the best agent team and sequencing.
+
+---
+
+## Cost-Aware Routing (v1.7.0)
+
+**What is cost-aware routing?**
+
+A router that matches your task to the right model tier (Haiku 4.5 for trivial work, Sonnet 4.6 for standard coding, Opus 4.6 for hard architecture).
+
+**How do I use it?**
+
+```bash
+npx kodelyth-ecc route "write a React component"
+# Suggests: Haiku 4.5 (fast, cheap, capable for components)
+
+npx kodelyth-ecc route "architecture a distributed payment system"
+# Suggests: Opus 4.6 (deep reasoning, long-term trade-offs)
+```
+
+**Does it help save money?**
+
+Yes. Most coding work runs fine on Sonnet (and often Haiku). The router prevents overspending on Opus for trivial tasks.
+
+---
+
+## SLSA L3 + SBOM (v1.7.0)
+
+**What is SLSA L3 support?**
+
+Software Artifact Signing Level 3 — ECC can generate and verify supply chain artifacts with cryptographic signatures.
+
+**How do I generate SBOM?**
+
+```bash
+npx kodelyth-ecc sbom
+```
+
+Generates Software Bill of Materials in SPDX format.
+
+**How do I create the manifest?**
+
+```bash
+npx kodelyth-ecc manifest
+```
+
+Creates a content manifest with file hashes and timestamps.
+
+**How do I verify integrity?**
+
+```bash
+npx kodelyth-ecc verify
+```
+
+Verifies SBOM, manifest, and signatures. Reports any tampering or missing artifacts.
+
+---
+
 ## Agents
 
 **Do I need to type agent names?**
 
 No. Semantic intent routing detects what you need from how you describe the problem. You never need to type `use debug-detective` — just describe the bug.
 
-**What are the 16 Kodelyth Exclusive agents?**
+**What are the 70 agents?**
 
-Agents that exist only in ECC and nowhere else: `kodelyth-advisor`, `kodelyth-memory`, `pair-programmer`, `debug-detective`, `silent-failure-hunter`, `incident-commander`, `load-tester`, `ux-reviewer`, `api-guardian`, `migration-guide`, `dependency-doctor`, `git-rescue`, `release-captain`, `env-debugger`, `flake-hunter`, `image-architect`.
+62 standard specialist agents (reviewers, debuggers, architects, etc.) + 8 adversarial devil-mode agents (injection hunter, supply chain auditor, secret hunter, etc.).
 
 **How do I invoke an agent explicitly?**
 
@@ -127,6 +294,12 @@ No. All hooks are async with aggressive timeouts. They fire in the background an
 
 API keys (OpenAI, Anthropic, AWS, GCP, Stripe, etc.), passwords and tokens in code, private keys (RSA, EC), and `.env` files accidentally staged. It blocks the commit and reports the exact line.
 
+**Are there safety hooks?**
+
+Yes. v1.7.0 adds:
+- `prompt-injection-guard` — Detects attempted prompt injection in inputs
+- `token-budget-enforcer` — Warns when approaching token limit per session
+
 ---
 
 ## Costs
@@ -149,7 +322,8 @@ cat ~/.claude/logs/token-usage.jsonl | jq -s 'sort_by(.tokens) | reverse | .[0:1
 1. Disable hooks you don't use: `export ECC_DISABLED_HOOKS=...`
 2. Keep `tasks/lessons.md` concise — long lesson files inject more tokens every session
 3. Use `/compact` (Claude Code) before large sessions to summarize context
-4. Match model to task — most work runs fine on Sonnet, not Opus
+4. Use `npx kodelyth-ecc route` to match model to task
+5. Review memory periodically — delete old or obsolete solutions
 
 ---
 
@@ -157,7 +331,7 @@ cat ~/.claude/logs/token-usage.jsonl | jq -s 'sort_by(.tokens) | reverse | .[0:1
 
 **Does ECC phone home?**
 
-Never. No telemetry, no analytics, no usage reporting of any kind.
+Never. No telemetry, no analytics, no usage reporting of any kind. The MCP server runs on localhost only.
 
 **Where is my data stored?**
 
@@ -171,6 +345,10 @@ Entirely on your machine:
 
 Yes. After install, ECC requires no internet access. The AI models themselves (Claude API, Gemini, etc.) still require connectivity — that is your AI tool, not ECC.
 
+**Is memory encrypted?**
+
+Memory files are plain text JSONL. For security-sensitive projects, you can encrypt `~/.kodelyth/memory/` and `tasks/lessons.md` with your OS or disk encryption. ECC has no built-in encryption (by design — simpler, more transparent).
+
 ---
 
 ## Troubleshooting
@@ -198,6 +376,23 @@ Check that `hooks/memory/auto-recall.js` is running. Verify in `~/.claude/settin
 
 On Claude Code, `image-architect` requires the fal.ai MCP server to be configured. If not available, it falls back to SVG generation. On Antigravity and Codex CLI, image generation is native and requires no configuration.
 
+**MCP server fails to start**
+
+Check that the port is not in use:
+```bash
+lsof -i :5173
+# Kill if needed: kill -9 <PID>
+npx kodelyth-ecc mcp --port 5174  # Use a different port
+```
+
+**Dashboard shows no sessions**
+
+Sessions are recorded when hooks are active (Claude Code only). On other platforms, run `/sessions` in your AI tool to see local session metadata.
+
+**devil-mode returns no findings**
+
+This is normal — not all code has vulnerabilities. devil-mode ran cleanly. If you suspect it should have found issues, try `/devil-mode --all` to run all 8 agents (slower but more thorough).
+
 ---
 
 ## Getting Help
@@ -205,3 +400,5 @@ On Claude Code, `image-architect` requires the fal.ai MCP server to be configure
 - GitHub Issues: [github.com/sifxprime/kodelyth-ecc/issues](https://github.com/sifxprime/kodelyth-ecc/issues)
 - Start a session and type `/kodelyth-quickstart` for an interactive tour
 - Type "I'm lost" or "I don't know where to start" — the `kodelyth-advisor` agent will orient you
+- For MCP server help: `npx kodelyth-ecc mcp --help`
+- For dashboard help: `npx kodelyth-ecc dashboard --help`

package/wiki/Home.md

@@ -1,8 +1,8 @@
 # Kodelyth ECC Wiki
 
-**Kodelyth Enhanced Coding Companion** — production-grade AI coding toolkit for Claude Code, Windsurf, Cursor, Codex CLI, Google Antigravity, and OpenCode.
+**Kodelyth Enhanced Coding Companion** — production-grade AI coding toolkit for Claude Code, Windsurf, Cursor, Codex CLI, Google Antigravity, and more.
 
-> Version: **v1.5.3** · Agents: **62** · Skills: **188** · Commands: **80** · Hooks: **20+** · Platforms: **6**
+> Version: **v1.7.0** · Agents: **70** (including 8 adversarial) · Skills: **194** · Commands: **97** (including 8 parallel) · Hooks: **22+** · Platforms: **12** · Tests: **336**
 
 ---
 
@@ -33,8 +33,11 @@ npx kodelyth-ecc
 npx kodelyth-ecc --target windsurf-project
 npx kodelyth-ecc --target antigravity
 npx kodelyth-ecc --target cursor-project
-npx kodelyth-ecc --target codex-home
-npx kodelyth-ecc --target opencode
+
+# Power bundles
+npx kodelyth-ecc --bundle indie-hacker
+npx kodelyth-ecc --bundle red-team
+npx kodelyth-ecc --bundle enterprise
 ```
 
 After install, type `/kodelyth-quickstart` in your AI tool to begin.
@@ -45,15 +48,17 @@ After install, type `/kodelyth-quickstart` in your AI tool to begin.
 
 | Component | Count | Description |
 |---|---|---|
-| Agents | **62** | Specialist subagents — reviewers, debuggers, architects, memory, image-architect |
-| Skills | **188** | Domain knowledge — patterns, testing, security, DevOps, memory protocol |
-| Commands | **80** | Slash workflows (`/tdd`, `/plan`, `/memory`, `/lessons`, `/project-launch`, ...) |
-| Hooks | **20+** | Quality gates, memory inject + capture, correction encoding, secret scanning |
-| Rules | **17** | Always-on standards + semantic routing + memory protocol + self-improvement |
+| Agents | **70** | Specialist subagents (62 standard + 8 adversarial devil-mode agents) |
+| Skills | **194** | Domain knowledge — patterns, testing, security, DevOps, memory, swarm orchestration |
+| Commands | **97** | Slash workflows (88 base + 8 parallel multi-agent + `/devil-mode` + utilities) |
+| Hooks | **20+** | Quality gates, memory inject + capture, correction encoding, safety guards |
+| Rules | **14** | Always-on standards, semantic routing, memory protocol, self-improvement |
 
 ---
 
-## Semantic Intent Routing (v1.5.3)
+## Key Features (v1.7.0)
+
+### Semantic Intent Routing
 
 The AI reads **intent behind words**, not just keywords. It reasons about emotion, context, and even what you paste.
 
@@ -69,11 +74,70 @@ The AI reads **intent behind words**, not just keywords. It reasons about emotio
 | [paste stack trace with no text] | `debug-detective` |
 | "Build failed on Vercel" | `build-error-resolver` |
 | "production is down, 500s everywhere" | `incident-commander` |
-| "will this hold under 10k users?" | `load-tester` |
+| "Find vulnerabilities in my code" | `/devil-mode` |
 
----
+### Parallel Commands (8 total)
+
+Fire multiple specialist agents at once and get results in 10-15 minutes instead of 45-60:
+
+| Command | Agents fired | Time saving |
+|---|---|---|
+| `/project-launch` | architect + pair-programmer + security-reviewer + tdd-guide + ux-reviewer | 45 min → 10 min |
+| `/team-review` | code-reviewer + security-reviewer + performance-optimizer + api-guardian | 60 min → 15 min |
+| `/security-audit` | security-reviewer + dependency-doctor + api-guardian | 30 min → 8 min |
+| `/debug-blitz` | debug-detective + silent-failure-hunter + env-debugger | 30 min → 8 min |
+| `/refactor-sprint` | refactor-cleaner + code-simplifier + type-design-analyzer + tdd-guide | 45 min → 12 min |
+| `/pre-release` | release-captain + security-reviewer + code-reviewer | 30 min → 8 min |
+| `/onboard` | code-explorer + architect + doc-updater | 45 min → 10 min |
+| `/devil-mode` | prompt-injection-hunter + supply-chain-auditor + secret-hunter + license-violation-finder (or all 8 with `--all`) | 60 min → 15 min |
+
+### Adversarial Devil-Mode Agents (v1.7.0 new)
+
+Dedicated red-team mode fires 4-8 specialized adversarial agents to find vulnerabilities your team missed:
+
+```bash
+/devil-mode                 # Fires top 4: injection, supply chain, secrets, license
+/devil-mode --all          # Fires all 8 (adds: jailbreak, code theft, backdoors, chaos)
+```
+
+The 8 devil-mode agents:
+- `prompt-injection-hunter` — Find AI injection vulnerabilities
+- `supply-chain-auditor` — Audit dependencies and build artifacts
+- `secret-hunter` — Hunt hardcoded secrets and API keys
+- `license-violation-finder` — Detect GPL/license compliance issues
+- `jailbreak-tester` — Test safety bypass attempts
+- `code-stealer-detector` — Find exfiltration vectors
+- `backdoor-hunter` — Identify inserted vulnerabilities
+- `chaos-engineer` — Failure mode and edge case analysis
 
-## Compound Learning System (v1.5.1)
+### MCP Server & Local Tooling
+
+```bash
+# Start the MCP server (16 tools, 6 prompts, 370+ resources)
+npx kodelyth-ecc mcp
+
+# Launch the local dashboard (5 tabs: Overview, Memory, Evolve, Catalog, Sessions)
+npx kodelyth-ecc dashboard
+
+# Cost-aware model routing
+npx kodelyth-ecc route "<task>"
+
+# Swarm orchestrator for parallel work
+npx kodelyth-ecc swarm --task "..."
+
+# Self-evolving memory system
+npx kodelyth-ecc evolve analyze
+
+# Session replay
+npx kodelyth-ecc replay <bundle>
+
+# SLSA L3 + SBOM verification
+npx kodelyth-ecc sbom
+npx kodelyth-ecc manifest
+npx kodelyth-ecc verify
+```
+
+### Compound Learning System
 
 Every correction you give Claude gets encoded permanently. The toolkit gets smarter every session.
 
@@ -85,8 +149,7 @@ Session 2:  read-lessons.js fires at session start
             → "PROJECT LESSONS — use pnpm not npm" injected
             → Claude uses pnpm without being told
 
-Month 3:    You try another tool. It uses npm. It asks basic questions.
-            You come back.
+Month 3:    You try another tool. It uses pnpm. You praise the consistency.
 ```
 
 **Three layers working together:**
@@ -95,28 +158,39 @@ Month 3:    You try another tool. It uses npm. It asks basic questions.
 |---|---|---|
 | Project Lessons | `tasks/lessons.md` | Per-project hard rules from your corrections |
 | Global Memory | `~/.kodelyth/memory/` | Cross-project BM25 recall of past solutions |
-| Semantic Routing | 62 agents | Routes every message to the right specialist |
+| Semantic Routing | 70 agents | Routes every message to the right specialist |
 
 ---
 
-## Parallel Commands (v1.5.2)
+## Platform Support (12 platforms)
 
-| Command | Fires | Time |
+Kodelyth ECC works across 12 AI coding platforms:
+
+| Platform | Status | Notes |
 |---|---|---|
-| `/project-launch` | architect + pair-programmer + security-reviewer + tdd-guide + ux-reviewer | 45 min → 10 min |
-| `/team-review` | code-reviewer + security-reviewer + performance-optimizer + api-guardian | 60 min → 15 min |
-| `/lessons` | Load/save/manage project lessons on all 6 platforms | Always works |
+| Claude Code | Full | All features, hooks auto-wired |
+| Windsurf (project) | Full | Agents, skills, commands, manual lessons |
+| Windsurf (home) | Full | Global installation |
+| Cursor (project) | Full | Agents, skills, commands, manual lessons |
+| Google Antigravity | Full | Native Gemini image gen, deep reasoning models |
+| Codex CLI (home) | Full | DALL-E 3 image gen native |
+| OpenCode | Full | Agents, skills, commands |
+| Cline | Full | VS Code extension integration |
+| Roocode | Full | Web IDE support |
+| Aider | Full | Git-native coding assistant |
+| Kimi | Full | Chinese AI platform |
+| Gemini (project/home) | Full | Dual-entry platform support |
 
 ---
 
 ## Wiki Pages
 
-- [Installation Guide](Installation-Guide) — all install methods, targets, profiles, OS support
-- [Agent Reference](Agent-Reference) — all 62 agents with descriptions and trigger patterns
-- [Skill Reference](Skill-Reference) — all major skills and how to load them
-- [Hook Reference](Hook-Reference) — all 20+ hooks and how they work
-- [Platform Support](Platform-Support) — per-platform capabilities and file layout
-- [FAQ](FAQ) — memory system, privacy, cost, troubleshooting
+- [Installation Guide](Installation-Guide) — all install methods, targets, profiles, bundles, OS support
+- [Agent Reference](Agent-Reference) — all 70 agents including 8 adversarial devil-mode agents
+- [Skill Reference](Skill-Reference) — all 194 skills including MCP, swarm, dashboard, and safety hooks
+- [Hook Reference](Hook-Reference) — all 22+ hooks and how they work
+- [Platform Support](Platform-Support) — per-platform capabilities for 12 platforms
+- [FAQ](FAQ) — memory, MCP server, dashboard, devil-mode, swarm, SLSA/SBOM, privacy, troubleshooting
 
 ---
 
@@ -124,7 +198,9 @@ Month 3:    You try another tool. It uses npm. It asks basic questions.
 
 | Version | Headline |
 |---|---|
-| **v1.5.3** | Semantic intent routing — intent not keywords, code/trace paste detection, 2–3× more patterns |
+| **v1.7.0** | Adversarial devil-mode (8 agents), MCP server, local dashboard, swarm orchestrator, cost-aware routing, self-evolving memory, SLSA L3 + SBOM, 12 platforms, 194 skills |
+| **v1.6.0** | 8 adversarial devil-mode agents, power bundles, 6 new IDE platform targets, 97 commands |
+| **v1.5.3** | Semantic intent routing — intent not keywords, code/trace paste detection, 62 agents, 188 skills |
 | **v1.5.2** | Parallel agents, `image-architect`, `/project-launch`, `/team-review`, `/lessons` |
 | **v1.5.1** | Compound learning — `capture-correction.js`, `read-lessons.js`, self-improvement workflow |
 | **v1.5.0** | `incident-commander`, `load-tester`, visual refresh, wiki rewrite |

package/wiki/Hook-Reference.md

@@ -1,6 +1,6 @@
 # Hook Reference
 
-ECC ships 20+ hooks that run automatically in the background — zero configuration after install.
+ECC ships 22+ hooks that run automatically in the background — zero configuration after install.
 
 Hooks are defined in `hooks/hooks.json` and wired into `~/.claude/settings.json` by the install script (Claude Code target only). All hooks are async and have aggressive timeouts to never block tool execution.
 

package/wiki/Installation-Guide.md

@@ -19,7 +19,7 @@ This detects your project type, installs the right component set, and wires ever
 
 ---
 
-## Targets
+## Targets (12 platforms)
 
 Install into a specific AI tool:
 
@@ -29,6 +29,7 @@ npx kodelyth-ecc
 
 # Windsurf
 npx kodelyth-ecc --target windsurf-project
+npx kodelyth-ecc --target windsurf-home
 
 # Cursor
 npx kodelyth-ecc --target cursor-project
@@ -41,19 +42,54 @@ npx kodelyth-ecc --target antigravity
 
 # OpenCode
 npx kodelyth-ecc --target opencode
+
+# Cline (VS Code extension)
+npx kodelyth-ecc --target cline
+
+# Roocode (web IDE)
+npx kodelyth-ecc --target roocode
+
+# Aider (git-native)
+npx kodelyth-ecc --target aider
+
+# Kimi (Chinese platform)
+npx kodelyth-ecc --target kimi
+
+# Gemini
+npx kodelyth-ecc --target gemini-project
+npx kodelyth-ecc --target gemini-home
 ```
 
 ---
 
+## Power Bundles (v1.7.0)
+
+Pre-configured agent + skill bundles for specific use cases:
+
+```bash
+# Indie hacker — web dev, rapid prototyping, solopreneur
+npx kodelyth-ecc --bundle indie-hacker
+
+# Red team — security audits, adversarial testing, pentesting
+npx kodelyth-ecc --bundle red-team
+
+# Enterprise — compliance, supply chain, SLA tracking, incident management
+npx kodelyth-ecc --bundle enterprise
+```
+
+Each bundle installs only the agents and skills relevant to that workflow.
+
+---
+
 ## What Gets Installed
 
 | Component | Count | Location |
 |---|---|---|
-| Agents | 62 | `~/.claude/agents/` (Claude Code) or project-local |
-| Skills | 188 | `~/.claude/skills/` or project-local |
-| Commands | 80 | `~/.claude/commands/` or project-local |
+| Agents | 70 | `~/.claude/agents/` (Claude Code) or project-local |
+| Skills | 194 | `~/.claude/skills/` or project-local |
+| Commands | 97 | `~/.claude/commands/` or project-local |
 | Hooks | 20+ | `~/.claude/settings.json` (Claude Code only) |
-| Rules | 17 | `~/.claude/rules/` or project-local |
+| Rules | 14 | `~/.claude/rules/` or project-local |
 
 ---
 
@@ -119,7 +155,7 @@ cd kodelyth-ecc
 
 ### 3. Verify
 
-Open Claude Code (or your AI tool) and type:
+Open your AI tool and type:
 ```
 /kodelyth-quickstart
 ```
@@ -134,6 +170,37 @@ Type `/kodelyth-quickstart` in your AI tool to begin. It walks you through agent
 
 ---
 
+## Quick Tooling Commands
+
+After install, you have access to local utilities:
+
+```bash
+# Start the MCP server (16 tools, 6 prompts, 370+ resources)
+npx kodelyth-ecc mcp
+
+# Launch the local dashboard (5 tabs: Overview, Memory, Evolve, Catalog, Sessions)
+npx kodelyth-ecc dashboard
+
+# Cost-aware model routing
+npx kodelyth-ecc route "<task>"
+
+# Swarm orchestrator for parallel work
+npx kodelyth-ecc swarm --task "..."
+
+# Self-evolving memory analysis
+npx kodelyth-ecc evolve analyze
+
+# Session replay
+npx kodelyth-ecc replay <bundle>
+
+# SLSA L3 + SBOM generation
+npx kodelyth-ecc sbom
+npx kodelyth-ecc manifest
+npx kodelyth-ecc verify
+```
+
+---
+
 ## Updating
 
 ```bash
@@ -170,6 +237,10 @@ To disable specific hooks:
 export ECC_DISABLED_HOOKS=kodelyth:prompt:recall,kodelyth:post-edit:test-reminder
 ```
 
+**Safety hooks (v1.7.0):**
+- `prompt-injection-guard` — Detects attempted prompt injection
+- `token-budget-enforcer` — Warns when approaching context limit
+
 ---
 
 ## Troubleshooting
@@ -189,3 +260,35 @@ export ECC_DISABLED_HOOKS=kodelyth:prompt:recall,kodelyth:post-edit:test-reminde
 **Install fails on Windows**
 - Run PowerShell as Administrator
 - Or use `npx kodelyth-ecc` directly (no admin required)
+
+**MCP server won't start**
+- Ensure port 5173 (default) is not in use: `lsof -i :5173`
+- Try a different port: `npx kodelyth-ecc mcp --port 5174`
+
+**Dashboard cannot connect**
+- Verify the MCP server is running: `npx kodelyth-ecc mcp`
+- Check firewall — localhost access should be allowed
+- Use `npx kodelyth-ecc dashboard --port 5175` to try a different port
+
+**Tests fail after install**
+
+```bash
+npm test
+```
+
+If tests fail, file an issue with the output. Most installs pass all 336 tests.
+
+---
+
+## Post-Install Checklist
+
+After installing, verify these work:
+
+- [ ] `/kodelyth-quickstart` loads in your AI tool
+- [ ] `/lessons` opens your project lessons
+- [ ] `/memory` shows your memory store
+- [ ] `npx kodelyth-ecc mcp` starts the server
+- [ ] `npx kodelyth-ecc dashboard` opens the dashboard
+- [ ] `npm test` passes all 336 tests (Claude Code only)
+
+If any step fails, re-run the installer or file an issue.