claude-raid 0.2.5 → 0.2.6

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (2) hide show
  1. package/README.md +190 -142
  2. package/package.json +1 -1
package/README.md CHANGED
@@ -27,78 +27,126 @@
27
27
  [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
28
28
  [![Zero Dependencies](https://img.shields.io/badge/dependencies-0-brightgreen.svg)](#)
29
29
  [![Node.js 18+](https://img.shields.io/badge/node-18%2B-blue.svg)](#prerequisites)
30
+ [![294 Tests](https://img.shields.io/badge/tests-294_passing-brightgreen.svg)](#)
30
31
 
31
- [Quick Start](#quick-start) • [How It Works](#how-it-works) • [The Team](#the-team) • [Configuration](#configuration) • [CLI Reference](#cli-reference)
32
+ [Quick Start](#quick-start) · [The Canonical Quest](#the-canonical-quest) · [The Party](#the-party) · [The Dungeon](#the-dungeon) · [Configuration](#configuration) · [CLI Reference](#cli-reference)
32
33
 
33
34
  </div>
34
35
 
35
36
  ---
36
37
 
38
+ ## What is Claude Raid?
39
+
40
+ Claude Raid turns a single Claude Code session into a **4-agent adversarial team** that designs, plans, builds, reviews, and ships code through structured phases. Instead of one AI guessing at a solution, four agents challenge each other's work until only battle-tested decisions survive.
41
+
42
+ The **Wizard** orchestrates. The **Warrior**, **Archer**, and **Rogue** each bring a different lens — stress tolerance, pattern coherence, and assumption destruction. They work in rounds, pin verified findings to a shared **Dungeon**, and no phase closes until the Wizard rules.
43
+
44
+ One command installs the system. One command starts a quest.
45
+
46
+ ---
47
+
37
48
  ## Quick Start
38
49
 
39
50
  ```bash
40
51
  npx claude-raid summon
41
52
  ```
42
53
 
43
- That's it. One command installs the full system into any project.
54
+ That's it. One command installs agents, hooks, skills, and config into your project's `.claude/` directory.
44
55
 
45
56
  ```bash
46
- # Preview what gets installed (no changes made)
57
+ # Preview what gets installed (no changes)
47
58
  npx claude-raid summon --dry-run
48
59
 
49
- # Start a raid
50
- tmux new-session -s raid
51
- claude --agent wizard
60
+ # Start a quest
61
+ claude-raid start
52
62
  ```
53
63
 
54
- Each agent gets its own tmux pane. Click into any pane to observe or talk to that agent directly. Describe your task and the Wizard takes over.
64
+ The Wizard greets you, you describe your task, and the quest begins.
55
65
 
56
66
  ### Prerequisites
57
67
 
58
- | Requirement | Why | Auto-configured? |
59
- |:--|:--|:--|
60
- | **Node.js** 18+ | Runs the installer | No |
61
- | **Claude Code** v2.1.32+ | Agent teams support | No |
62
- | **tmux** | Multi-pane agent display | No — `brew install tmux` |
63
- | **jq** | Hook config parsing | Pre-installed on macOS |
64
- | **teammateMode** | Set to `tmux` in `~/.claude.json` | Yes (wizard prompts) |
68
+ | Requirement | Why |
69
+ |:--|:--|
70
+ | **Node.js** 18+ | Runs the CLI |
71
+ | **Claude Code** v2.1.32+ | Agent teams support |
72
+ | **tmux** | Multi-pane agent display (`brew install tmux`) |
73
+ | **jq** | Config parsing (pre-installed on macOS) |
65
74
 
66
- > **tmux is required for the multi-pane experience.** Each agent runs in its own pane so you can observe and interact with them independently. Without tmux, agents run in-process (single pane, cycle with `Shift+Down`).
75
+ > **tmux** gives each agent its own pane click into any pane to observe or talk to that agent directly. Without tmux, agents run in-process and you cycle between them with `Shift+Down`.
67
76
 
68
77
  The setup wizard checks all prerequisites during `summon` and offers to fix what it can.
69
78
 
70
79
  ---
71
80
 
72
- ## How It Works
81
+ ## The Canonical Quest
82
+
83
+ The Canonical Quest is a 6-phase development cycle. Every feature, refactor, or system built through the Raid follows this sequence.
84
+
85
+ ```mermaid
86
+ flowchart LR
87
+ PRD["Phase 1\nPRD"]
88
+ Design["Phase 2\nDesign"]
89
+ Plan["Phase 3\nPlan"]
90
+ Impl["Phase 4\nImplementation"]
91
+ Review["Phase 5\nReview"]
92
+ Wrap["Phase 6\nWrap Up"]
93
+
94
+ PRD -->|"optional"| Design
95
+ Design --> Plan
96
+ Plan --> Impl
97
+ Impl --> Review
98
+ Review -->|"optional"| Wrap
99
+ Impl -->|"skip review"| Wrap
100
+
101
+ style PRD fill:#6c5ce7,color:#fff
102
+ style Design fill:#6c5ce7,color:#fff
103
+ style Plan fill:#6c5ce7,color:#fff
104
+ style Impl fill:#e17055,color:#fff
105
+ style Review fill:#00b894,color:#fff
106
+ style Wrap fill:#fdcb6e,color:#000
107
+ ```
73
108
 
74
- You describe a task. The Wizard assesses complexity, recommends a mode, and opens the **Dungeon** a shared artifact where agents pin verified findings throughout each phase.
109
+ ### Phase 1PRD *(optional)*
75
110
 
76
- ```
77
- DESIGN Agents explore the problem from competing angles.
78
- Challenge each other's findings. Pin what survives.
79
- Wizard closes when the design is battle-tested.
111
+ Agents research the problem space and produce a complete Product Requirements Document. No code. The Wizard mediates questions between agents and the human.
80
112
 
81
- PLAN Decompose the design into testable tasks.
82
- Fight over naming, ordering, coverage, compliance.
83
- Pin the agreed task list.
113
+ **Output:** `phase-1-prd.md`
84
114
 
85
- IMPLEMENTATION One agent builds each task (TDD enforced).
86
- The others attack the implementation directly.
87
- Every task earns approval before moving on.
115
+ ### Phase 2 Design
88
116
 
89
- REVIEW Independent reviews against design and plan.
90
- Agents fight over findings AND missing findings.
91
- Critical and Important issues must be fixed.
117
+ Agents explore design approaches from competing angles. Each brings their lens — the Warrior stress-tests architecture choices, the Archer traces ripple effects, the Rogue attacks assumptions. The design survives only if it withstands all three.
92
118
 
93
- FINISHING Agents debate completeness. Wizard presents
94
- merge options: merge, PR, keep branch, or discard.
95
- ```
119
+ **Output:** `phase-2-design.md` with mermaid diagrams and battle-tested decisions
120
+
121
+ ### Phase 3 — Plan
122
+
123
+ Agents decompose the approved design into discrete, testable tasks. They fight over ordering, scope boundaries, naming, and test coverage until the plan earns consensus.
96
124
 
97
- No phase is skipped. No work passes unchallenged.
125
+ **Output:** `phase-3-plan.md` + individual task files (`phase-3-plan-task-01.md`, etc.)
126
+
127
+ ### Phase 4 — Implementation
128
+
129
+ The Wizard assigns tasks in batches. One agent builds each task using strict **TDD** (RED-GREEN-REFACTOR). The others cross-test the implementation — reading code, running tests, and challenging decisions. Every task earns approval before the next batch starts.
130
+
131
+ **Output:** `phase-4-implementation.md` + committed, tested code
132
+
133
+ ### Phase 5 — Review *(optional)*
134
+
135
+ Two sub-phases: **Pinning** (find issues) and **Fixing** (resolve them). Agents review independently, then fight over findings *and* missing findings. Critical and Important issues must be fixed. The **Black Card** system handles breaking architectural concerns.
136
+
137
+ **Output:** `phase-5-review.md`
138
+
139
+ ### Phase 6 — Wrap Up
140
+
141
+ The Wizard generates a quest storyboard summarizing what was built and why. Creates the PR, archives the dungeon to the vault, and dismisses the party.
142
+
143
+ **Output:** `phase-6-wrap-up.md` + PR + vault archive
98
144
 
99
145
  ---
100
146
 
101
- ## The Team
147
+ ## The Party
148
+
149
+ Four agents, each with a distinct lens. They collaborate through rigor, not agreement.
102
150
 
103
151
  <table>
104
152
  <tr>
@@ -108,37 +156,43 @@ No phase is skipped. No work passes unchallenged.
108
156
  <td width="25%" align="center"><h3>Rogue</h3><em>Assumption Destroyer</em></td>
109
157
  </tr>
110
158
  <tr>
111
- <td valign="top">Thinks 5 times before speaking. Opens phases, dispatches the team, observes silently, and closes with binding rulings citing Dungeon evidence. <strong>Never writes code.</strong></td>
112
- <td valign="top">Race conditions, null input, scale, memory pressure nothing passes unchecked. Demands evidence, proposes counter-examples, pushes until things break. Concedes instantly when proven wrong.</td>
113
- <td valign="top">Finds what brute force misses. Naming mismatches, violated conventions, design drift. Traces ripple effects across modules through implicit contracts.</td>
114
- <td valign="top">Thinks like a malicious user, a failing network, a corrupted database. Constructs the exact sequence of events that turns a minor oversight into a critical failure.</td>
159
+ <td valign="top">Thinks 5 times before speaking. Opens phases, dispatches the team, observes, and closes with binding rulings. The bridge between agents, dungeon, and human. <strong>Never writes code.</strong></td>
160
+ <td valign="top">Does this hold under pressure? Tests boundaries, load, edge cases, and failure modes. Brings the exact scenario that breaks it not just "this is wrong."</td>
161
+ <td valign="top">Does this fit? Traces how changes ripple through the system. Catches naming drift, contract violations, and implicit dependencies that break silently.</td>
162
+ <td valign="top">What did everyone assume that isn't guaranteed? Thinks like a failing system, a malicious input, a race condition. Constructs the attack sequence that turns oversight into failure.</td>
115
163
  </tr>
116
164
  </table>
117
165
 
118
- ---
119
-
120
- ## Modes
121
-
122
- Not every task needs all four agents. The Wizard recommends a mode based on complexity — or you can override directly.
123
-
124
- | | Full Raid | Skirmish | Scout |
125
- |:--|:--|:--|:--|
126
- | **Agents** | 3 + Wizard | 2 + Wizard | 1 + Wizard |
127
- | **Design** | Full adversarial | Lightweight | Inline |
128
- | **Planning** | Full adversarial | Merged with design | Inline |
129
- | **Implementation** | 1 builds, 2 attack | 1 builds, 1 attacks | 1 builds, Wizard reviews |
130
- | **Review** | 3 independent reviews | 1 review + Wizard | Wizard only |
131
- | **TDD** | Enforced | Enforced | Enforced |
166
+ ### How They Work Together
167
+
168
+ ```mermaid
169
+ sequenceDiagram
170
+ participant W as Wizard
171
+ participant Party as Warrior / Archer / Rogue
172
+
173
+ W->>Party: Dispatch tasks + angles
174
+ Note over Party: Parallel independent work
175
+ Party->>W: ROUND_COMPLETE
176
+ W->>Party: Cross-test each other's work
177
+ Note over Party: Challenge, verify, concede or escalate
178
+ Party->>W: Findings + evidence
179
+ W->>W: Pin verified findings to Dungeon
180
+ Note over W: Next round or phase close
181
+ ```
132
182
 
133
- **When to use each:**
183
+ **Round-based, not real-time.** Agents work independently, flag completion, then cross-test. No mid-thinking interruptions. Each message carries evidence and a conclusion. Converge in 2-3 exchanges per finding — escalate to the Wizard after 3.
134
184
 
135
- - **Full Raid** — Architecture decisions, security-critical code, major refactors, cross-layer changes
136
- - **Skirmish** — Medium features, non-trivial bugfixes, multi-file changes
137
- - **Scout** — Config changes, documentation, single-file fixes
185
+ ### Seven Pillars
138
186
 
139
- Override the Wizard's recommendation: *"Full Raid this"*, *"Skirmish this bugfix"*, *"Scout this"*.
187
+ Every agent, every phase, every interaction:
140
188
 
141
- > The Wizard can escalate mid-task (Scout to Skirmish, Skirmish to Full Raid) with your approval. It cannot de-escalate without asking.
189
+ 1. **Intellectual Honesty** Every claim backed by evidence gathered this turn. No guessing.
190
+ 2. **Zero Ego** — Concede instantly when proven wrong. A teammate catching your mistake is a gift.
191
+ 3. **Discipline** — Every interaction carries work forward. If you're not adding information, stop talking.
192
+ 4. **Round-Based Interaction** — Turn-based work, flag completion, cross-test on dispatch.
193
+ 5. **Question Chain** — Agents never ask the human directly. All questions flow through the Wizard.
194
+ 6. **Phase Spoils** — Every phase produces a detailed markdown artifact. No exceptions.
195
+ 7. **Black Cards** — Architecture-breaking findings that cannot be fixed within the current design. Escalated to the human with rollback options.
142
196
 
143
197
  ---
144
198
 
@@ -146,11 +200,34 @@ Override the Wizard's recommendation: *"Full Raid this"*, *"Skirmish this bugfix
146
200
 
147
201
  The Dungeon is the team's shared knowledge artifact — a curated board where agents pin verified findings during each phase.
148
202
 
149
- **What goes in:** Findings that survived challenge from 2+ agents, active unresolved battles, key decisions, escalation points.
203
+ | Concept | What It Means |
204
+ |:--|:--|
205
+ | **Quest** | A complete session — from greeting to PR |
206
+ | **Dungeon** | The quest's artifact directory: phase files, task files, findings |
207
+ | **Vault** | Archive of completed quests for institutional memory |
208
+ | **Phase Spoils** | Mandatory output of each phase: a detailed markdown report |
209
+ | **Black Card** | A finding that fundamentally breaks the architecture — requires human decision |
210
+ | **Pin** | A verified finding that survived challenge from 2+ agents |
150
211
 
151
- **What stays in conversation:** The back-and-forth of challenges, exploratory thinking, concessions. The conversation is the sparring ring. The Dungeon is the scoreboard.
212
+ ### Dungeon Filesystem
213
+
214
+ ```
215
+ .claude/dungeon/{quest-slug}/ # Active quest artifacts
216
+ ├── phase-1-prd.md # PRD (optional)
217
+ ├── phase-2-design.md # Battle-tested design
218
+ ├── phase-3-plan.md # Task index
219
+ ├── phase-3-plan-task-01.md # Individual task specs
220
+ ├── phase-4-implementation.md # Implementation log
221
+ ├── phase-5-review.md # Review board (optional)
222
+ └── phase-6-wrap-up.md # Quest storyboard
223
+
224
+ .claude/vault/{quest-slug}/ # Archived completed quests
225
+ .claude/raid-session # Active session state
226
+ ```
152
227
 
153
- **Lifecycle:** The Wizard creates it when opening a phase. Agents pin findings with `DUNGEON:` prefix. The Wizard archives it when closing. All Dungeon files are cleaned up when the session ends.
228
+ **What goes in the Dungeon:** Findings that survived challenge from 2+ agents, key decisions, escalation points.
229
+
230
+ **What stays in conversation:** Back-and-forth challenges, exploratory thinking, concessions. The conversation is the sparring ring. The Dungeon is the scoreboard.
154
231
 
155
232
  ---
156
233
 
@@ -172,12 +249,12 @@ The Dungeon is the team's shared knowledge artifact — a curated board where ag
172
249
  │ └── rogue.md # Assumption destroyer
173
250
  ├── hooks/
174
251
  │ ├── raid-lib.sh # Shared config and session state
175
- │ ├── raid-session-start.sh # Session activation
176
- │ ├── raid-session-end.sh # Archive and cleanup
177
- │ ├── raid-pre-compact.sh # Pre-compaction backup
252
+ │ ├── raid-session-start.sh # Session activation + quest directory
253
+ │ ├── raid-session-end.sh # Archive to vault + cleanup
254
+ │ ├── raid-pre-compact.sh # Pre-compaction dungeon backup
178
255
  │ ├── raid-task-created.sh # Task subject validation
179
- │ ├── validate-commit.sh # Conventional commits + test gate
180
- │ ├── validate-write-gate.sh # Design doc before implementation
256
+ │ ├── validate-commit.sh # Conventional commits
257
+ │ ├── validate-write-gate.sh # Phase-based write protection
181
258
  │ ├── validate-file-naming.sh # Naming convention enforcement
182
259
  │ ├── validate-no-placeholders.sh # No TBD/TODO in specs
183
260
  │ ├── validate-dungeon.sh # Multi-agent verification on pins
@@ -191,82 +268,40 @@ The Dungeon is the team's shared knowledge artifact — a curated board where ag
191
268
  ├── raid-canonical-implementation-plan/ # Phase 3: Task decomposition
192
269
  ├── raid-canonical-implementation/ # Phase 4: TDD + cross-testing
193
270
  ├── raid-canonical-review/ # Phase 5: Pinning + fixing
194
- ├── raid-wrap-up/ # Phase 6: Storyboard + PR
271
+ ├── raid-wrap-up/ # Phase 6: Storyboard + PR + vault
195
272
  ├── raid-tdd/ # RED-GREEN-REFACTOR enforcement
196
273
  ├── raid-verification/ # Evidence-before-claims gate
197
274
  ├── raid-debugging/ # Root-cause investigation
198
275
  ├── raid-browser/ # Browser orchestration
199
- ├── raid-browser-playwright/ # Playwright test authoring
200
276
  └── raid-browser-chrome/ # Live Chrome inspection
201
277
  ```
202
278
 
203
279
  </details>
204
280
 
205
- Runtime files (dungeon artifacts, session state) are created during Raid sessions and cleaned up automatically.
206
-
207
- ---
208
-
209
- ## Hooks
210
-
211
- Hooks enforce workflow discipline automatically. They only activate during Raid sessions — they never interfere with normal coding.
212
-
213
- ### Lifecycle
281
+ ### Agents (4)
214
282
 
215
- | Hook | Trigger | Purpose |
216
- |:--|:--|:--|
217
- | `raid-session-start` | SessionStart | Activates Raid workflow, checks Vault for past quests |
218
- | `raid-session-end` | SessionEnd | Archives Dungeon, drafts Vault entry, removes session |
219
- | `raid-pre-compact` | PreCompact | Backs up Dungeon before message compaction |
220
- | `raid-task-created` | TaskCreated | Validates task subjects are meaningful |
221
-
222
- ### Quality Gates
283
+ The Wizard orchestrates. Warrior, Archer, and Rogue each bring a specialized lens. All agents run on Claude Opus 4.6.
223
284
 
224
- | Hook | Trigger | Purpose |
225
- |:--|:--|:--|
226
- | `validate-commit` | PreToolUse (Bash) | Conventional commit format + test gate |
227
- | `validate-write-gate` | PreToolUse (Write) | Blocks implementation before design doc exists |
228
- | `validate-file-naming` | PostToolUse (Write) | Enforces naming conventions |
229
- | `validate-no-placeholders` | PostToolUse (Write) | Blocks TBD/TODO/FIXME in specs and plans |
230
- | `validate-dungeon` | PostToolUse (Write) | Requires 2+ agents verified on Dungeon pins |
231
- | `validate-browser-tests-exist` | PreToolUse (Bash) | Checks Playwright tests exist before commits |
232
- | `validate-browser-cleanup` | PostToolUse (Bash) | Verifies browser processes cleaned up |
285
+ ### Hooks (12)
233
286
 
234
- All hooks are POSIX-compatible, read config from `raid.json`, and use `#claude-raid` markers to avoid collisions with your existing hooks. Exit code `0` = pass, `2` = block with message.
287
+ Hooks enforce workflow discipline automatically and **only activate during Raid sessions** they never interfere with normal coding.
235
288
 
236
- ---
289
+ **Lifecycle hooks** manage session start/end, quest directory creation, vault archival, and context compaction backup.
237
290
 
238
- ## Skills
291
+ **Quality gate hooks** enforce conventional commits, phase-based write protection, naming conventions, placeholder blocking, multi-agent verification on dungeon pins, and browser test detection.
239
292
 
240
- 14 specialized skills guide agent behavior across the workflow.
293
+ All hooks are POSIX-compatible and use `#claude-raid` markers to coexist safely with your existing hooks.
241
294
 
242
- ### Phase Skills
243
-
244
- | Skill | Purpose |
245
- |:--|:--|
246
- | `raid-init` | Quest selection, greeting, session setup |
247
- | `raid-canonical-protocol` | Canonical Quest rules, signals, phase gates |
248
- | `raid-canonical-prd` | Phase 1 — PRD creation (optional) |
249
- | `raid-canonical-design` | Phase 2 — Adversarial design exploration |
250
- | `raid-canonical-implementation-plan` | Phase 3 — Task decomposition |
251
- | `raid-canonical-implementation` | Phase 4 — TDD implementation + cross-testing |
252
- | `raid-canonical-review` | Phase 5 — Pinning, fixing, black cards |
253
- | `raid-wrap-up` | Phase 6 — Storyboard, PR, vault archival |
254
-
255
- ### Discipline Skills
256
-
257
- | Skill | Purpose |
258
- |:--|:--|
259
- | `raid-tdd` | Strict RED-GREEN-REFACTOR. No production code before a failing test. |
260
- | `raid-debugging` | Competing hypotheses in parallel. No fixes without confirmed root cause. |
261
- | `raid-verification` | Evidence before assertions. Fresh test run required before any completion claim. |
295
+ ### Skills (13)
262
296
 
263
- ### Browser Skills
297
+ Skills guide agent behavior across the workflow. Three categories:
264
298
 
265
- | Skill | Purpose |
266
- |:--|:--|
267
- | `raid-browser` | Browser startup discovery — dev server, auth, startup steps |
268
- | `raid-browser-playwright` | Playwright MCP test authoring with network and console assertions |
269
- | `raid-browser-chrome` | Live browser inspection via Claude-in-Chrome MCP |
299
+ | Category | Skills | Purpose |
300
+ |:--|:--|:--|
301
+ | **Core** | `raid-init` | Quest selection, greeting, session bootstrap |
302
+ | **Canonical Quest** | 7 phase skills | One skill per phase, chained in order |
303
+ | **Discipline** | `raid-tdd`, `raid-verification`, `raid-debugging` | Quest-agnostic enforcement invoked within any phase |
304
+ | **Browser** | `raid-browser`, `raid-browser-chrome` | Browser orchestration and live inspection |
270
305
 
271
306
  ---
272
307
 
@@ -307,13 +342,13 @@ All hooks are POSIX-compatible, read config from `raid.json`, and use `#claude-r
307
342
 
308
343
  | Marker File | Language | Test Command | Lint Command |
309
344
  |:--|:--|:--|:--|
310
- | `package.json` | JavaScript | `npm test` | `npm run lint` |
345
+ | `package.json` | JavaScript/TypeScript | `npm test` | `npm run lint` |
311
346
  | `Cargo.toml` | Rust | `cargo test` | `cargo clippy` |
312
347
  | `pyproject.toml` | Python | `pytest` / `poetry run pytest` | `ruff check .` |
313
348
  | `requirements.txt` | Python | `pytest` | `ruff check .` |
314
349
  | `go.mod` | Go | `go test ./...` | `go vet ./...` |
315
350
 
316
- Commands are extracted from your project files (e.g., `scripts.test` in `package.json`). Package manager is auto-detected (npm, pnpm, yarn, bun, uv, poetry). Edit `raid.json` to override any value.
351
+ Package manager is auto-detected (npm, pnpm, yarn, bun, uv, poetry). Commands are extracted from your project files where possible. Edit `raid.json` to override any value.
317
352
 
318
353
  <details>
319
354
  <summary><strong>Full configuration reference</strong></summary>
@@ -330,7 +365,6 @@ Commands are extracted from your project files (e.g., `scripts.test` in `package
330
365
  | `conventions.commits` | `conventional` | Commit message format |
331
366
  | `conventions.commitMinLength` | `15` | Minimum commit message length |
332
367
  | `conventions.maxDepth` | `8` | Maximum file nesting depth |
333
- | `raid.defaultMode` | `full` | Default mode: `full`, `skirmish`, `scout` |
334
368
  | `raid.lifecycle.testWindowMinutes` | `10` | Max age (minutes) of test run for verification |
335
369
 
336
370
  </details>
@@ -352,7 +386,7 @@ When a browser framework is detected (Next.js, Vite, Angular, etc.), a `browser`
352
386
  }
353
387
  ```
354
388
 
355
- This enables browser-specific hooks and skills — Playwright test detection, browser process cleanup, and live inspection during reviews.
389
+ This enables browser-specific hooks and skills — Playwright test detection, browser process cleanup, and live Chrome inspection during reviews.
356
390
 
357
391
  ---
358
392
 
@@ -360,14 +394,20 @@ This enables browser-specific hooks and skills — Playwright test detection, br
360
394
 
361
395
  | Command | Alias | Purpose |
362
396
  |:--|:--|:--|
397
+ | `claude-raid start` | — | Launch the Wizard and begin a quest |
363
398
  | `claude-raid summon` | `init` | Install Raid into your project |
364
399
  | `claude-raid update` | — | Upgrade hooks, skills, and rules to latest |
365
400
  | `claude-raid dismantle` | `remove` | Remove all Raid files, restore original settings |
366
401
  | `claude-raid heal` | `doctor` | Check environment health |
402
+ | `claude-raid sync` | — | Git pull + re-summon |
367
403
 
368
404
  <details>
369
405
  <summary><strong>Command details</strong></summary>
370
406
 
407
+ ### `start`
408
+
409
+ Launches `claude --agent wizard` with full permissions. The Wizard loads its rules, greets you, and begins quest selection. This is the primary entry point for running a quest.
410
+
371
411
  ### `summon`
372
412
 
373
413
  Installs the full Raid system. Auto-detects project type, copies agents/hooks/skills, generates `raid.json`, merges settings (with backup), and runs the setup wizard.
@@ -378,7 +418,7 @@ Installs the full Raid system. Auto-detects project type, copies agents/hooks/sk
378
418
 
379
419
  ### `update`
380
420
 
381
- Upgrades hooks, skills, and rules to the latest version. Skips customized agents (warns which ones were preserved). Does not touch `raid.json`.
421
+ Upgrades hooks, skills, and rules to the latest version. Skips customized agents and warns which ones were preserved. Does not touch `raid.json`.
382
422
 
383
423
  ### `dismantle`
384
424
 
@@ -386,7 +426,11 @@ Removes all Raid agents, hooks, skills, and config files. Restores `settings.jso
386
426
 
387
427
  ### `heal`
388
428
 
389
- Checks Node.js, Claude Code, jq, teammateMode, and split-pane support. Offers to fix missing configuration interactively.
429
+ Checks Node.js, Claude Code, jq, tmux, and teammateMode. Offers to fix missing configuration interactively.
430
+
431
+ ### `sync`
432
+
433
+ Pulls latest from remote and re-runs summon to pick up any template changes. Useful after CI bumps the version.
390
434
 
391
435
  </details>
392
436
 
@@ -416,25 +460,29 @@ Checks Node.js, Claude Code, jq, teammateMode, and split-pane support. Offers to
416
460
  ## Design Principles
417
461
 
418
462
  - **Non-invasive** — Never touches your `CLAUDE.md`. Merges settings alongside your existing config with automatic backup. Clean removal restores originals.
419
- - **Session-scoped** — Quality gate hooks only activate during Raid sessions. They never interfere with normal coding.
463
+ - **Session-scoped** — Quality gate hooks only activate during Raid sessions. Normal coding is never affected.
420
464
  - **Zero dependencies** — Pure Node.js stdlib. Fast `npx` cold-start.
421
- - **Safe by default** — Never overwrites existing files. Customized agents are always preserved.
465
+ - **Safe by default** — Never overwrites existing files. Customized agents and rules are always preserved.
466
+ - **Round-based discipline** — Agents work in parallel, flag completion, then cross-test. No mid-thinking interruptions.
467
+ - **Question chain** — Agents never ask the human directly. All questions flow through the Wizard.
468
+ - **Wizard never implements** — Dispatches, observes, digests, rules. The party writes code.
469
+ - **Phase commits** — The Wizard commits at every phase transition with the quest name, phase, and summary.
422
470
 
423
471
  ---
424
472
 
425
- ## Inherited from Superpowers
473
+ ## Heritage
426
474
 
427
- Adapted from [obra/superpowers](https://github.com/obra/superpowers) by Jesse Vincent.
475
+ Adapted from [obra/superpowers](https://github.com/obra/superpowers) by Jesse Vincent. The Raid inherits five core enforcement principles:
428
476
 
429
- | Concept | How the Raid Uses It |
477
+ | Principle | How the Raid Enforces It |
430
478
  |:--|:--|
431
- | HARD-GATEs | No code before design approval. No implementation before plan approval. |
432
- | TDD Iron Law | No production code without a failing test first. Enforced in all modes. |
433
- | Verification Iron Law | No completion claims without fresh test evidence. |
434
- | No Placeholders | Specs and plans must contain complete content, not "TBD" or "implement later". |
435
- | Conventional Commits | Enforced via hook: `type(scope): description`. |
479
+ | **HARD-GATEs** | No code before design approval. No implementation before plan approval. |
480
+ | **TDD Iron Law** | No production code without a failing test first. Enforced in every phase. |
481
+ | **Verification Iron Law** | No completion claims without fresh test evidence. |
482
+ | **No Placeholders** | Specs and plans must contain complete content no "TBD" or "implement later". |
483
+ | **Conventional Commits** | Enforced via hook: `type(scope): description`. |
436
484
 
437
- **What's different:** Superpowers uses a single agent with subagent delegation. The Raid uses 4 persistent agents with adversarial cross-testing and direct interaction. Agents talk to each other, pin verified findings to a shared Dungeon, and self-organize within phases. Every decision is stress-tested from multiple angles before it passes.
485
+ **What's different:** Superpowers uses a single agent with subagent delegation. The Raid uses 4 persistent agents with adversarial cross-testing. Agents challenge each other directly, pin verified findings to a shared Dungeon, and self-organize within phases. Every decision is stress-tested from three angles before it passes.
438
486
 
439
487
  ---
440
488
 
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "claude-raid",
3
- "version": "0.2.5",
3
+ "version": "0.2.6",
4
4
  "type": "commonjs",
5
5
  "description": "Adversarial multi-agent development system for Claude Code",
6
6
  "author": "Pedro Picardi",