@ionivetech/mugiwara 0.5.0 → 0.5.2

package/.opencode/plugins/mugiwara.mjs

@@ -1,8 +1,9 @@
 // mugiwara — OpenCode plugin.
 //
-// Registers the Mugiwara crew (25 skills + 15 agents) with OpenCode from the
-// npm/git package, and announces the crew at session start. No runtime, no
-// deps — reads the markdown source of truth (content/) at config load.
+// Registers skills/agents via config hook (superpowers pattern: one plugin,
+// zero file-copy), announces the crew at session start, and handles runtime
+// mode switching. CLI `mugiwara install --target opencode` remains available
+// as an alternative file-based path.
 //
 // Install: add to opencode.json
 //   { "plugin": ["@ionivetech/mugiwara"] }
@@ -28,7 +29,7 @@ export function readMode({ projectDir = process.cwd(), home = homedir() } = {})
     for (const line of readFileSync(file, 'utf8').split(/\r?\n/)) {
       const t = line.trim();
       if (!t || t.startsWith('#')) continue;
-      const [k, v] = t.split('=').map((s) => s.trim());
+      const [k, v] = line.split('=').map((s) => s.trim());
       if (k !== 'mode') continue;
       return VALID_MODES.has(v) ? v : 'INVALID';
     }
@@ -50,8 +51,6 @@ const agentsDir = join(contentDir, 'agents');
 const ANNOUNCE =
   "Mugiwara crew available. The workflow auto-activates for non-trivial requests — no need to call `/using-mugiwara` at session start (it remains an optional router). Run the crew pipeline inline in the main conversation: embody ONE crew role at a time using its skill, wait for its report, then move to the next. Never Task-dispatch a crew member — the crew runs in the main thread; subagents only for [PARALLEL] task batches, concurrent review/security, and independent re-run checks. Progress shows as checkpoint reports at wave/stage boundaries, pausing on failure or risk. Switch mode with `/mugiwara-mode` (guided|semi|auto). See skills/mugiwara-workflow.";
 
-// OpenCode per-agent tuning. Content stays portable markdown; these knobs
-// (color, temperature, permission, steps) are opencode-only.
 const CREW = {
   'using-mugiwara': { color: '#84cc16', temperature: 0.2, steps: 10 },
   'luffy-orchestrator': { color: '#ef4444', temperature: 0.2, steps: 15 },
@@ -173,6 +172,8 @@ export function applyModeChange(mode, { projectDir = process.cwd(), home = homed
 }
 
 export default async () => ({
+  dispose: () => {},
+
   config: (config) => {
     config.skills = config.skills || {};
     config.skills.paths = config.skills.paths || [];
@@ -185,8 +186,6 @@ export default async () => ({
     }
   },
 
-  // Intercept user messages to detect `/mugiwara-mode` and natural-language
-  // mode toggles; write `.mugiwara/config` so the next wave reads the new level.
   'chat.message': async (_input, output) => {
     if (!output) return;
     if (typeof output === 'string') {
@@ -204,9 +203,6 @@ export default async () => ({
     }
   },
 
-  // ponytail-proven hook; appends the announce string to the system prompt.
-  // Dedupes: repeated transforms (model switch/compaction) must not grow the
-  // prompt unbounded — only append once per string content.
   'experimental.chat.system.transform': async (_input, output) => {
     if (!output.system.some((s) => s.includes('Mugiwara crew available'))) {
       if (output.system.length > 0) {

package/content/skills/mugiwara-context-budget/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: mugiwara-context-budget
-description: Use in large codebase, long session, or near context limit — feed selectively, trust-sort sources, progressive disclosure. The window is a budget, not a bin.
+description: Use in large codebase, long session, or near token limit — prioritize relevant files, trust-sort, progressive disclosure. Window is budget, not bin.
 ---
 
 # Context Engineering

package/content/skills/mugiwara-frontend/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: mugiwara-frontend
-description: Use for frontend implementation or redesign — matching repo standards, design-system extraction, responsive behavior, WCAG 2.1 AA. Bans AI-slop patterns.
+description: Use for frontend UI component, CSS styling, responsive layout, a11y/accessible — matching repo standards, design-system extraction, WCAG 2.1 AA. Bans AI-slop patterns.
 ---
 
 # Frontend (Anti-Slop)

package/content/skills/mugiwara-healing/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: mugiwara-healing
-description: Use when earlier waves produced failures — reads blocker ledger, stop-the-line triage, root-cause fixes, prove-it before fixing. Max 3 cycles.
+description: Use when earlier waves produced failures, broken things to fix — reads blocker ledger, stop-the-line triage, root-cause fixes, prove-it before fixing. Max 3 cycles.
 ---
 
 # Healing (Brook)

package/content/skills/mugiwara-orchestration/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: mugiwara-orchestration
-description: Use to triage a new mission — 5-way classification, lane sizing, wave coordination, check-ins, closure. Captain; never implements code.
+description: Use to triage a new mission — 5-way classify, coordinate waves, lane sizing, check-ins, closure. Captain; never implements code.
 ---
 
 # Orchestration (Luffy)

package/content/skills/mugiwara-planning/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: mugiwara-planning
-description: Use for turning an approved spec into an execution plan — interview-first, full context scan, scaled Quick/Standard/Full plans, parallel-proof waves.
+description: Use for turning an approved spec into an implementation plan — interview-first, full context scan, scaled Quick/Standard/Full plans, parallel-proof waves.
 ---
 
 # Planning (Nami)

package/content/skills/mugiwara-proof-order/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: mugiwara-proof-order
-description: Use when writing code — RED-GREEN-REFACTOR. Proof value from WHEN the test runs, not that it exists. One test = one behavior.
+description: Use when writing code — TDD, RED-GREEN-REFACTOR, test-driven. Proof value from WHEN the test runs, not that it exists. One test = one behavior.
 ---
 
 # Test-Driven Development

package/content/skills/mugiwara-quality/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: mugiwara-quality
-description: Use after checkpoint passes — formatter, linter, unit tests. Discovers project tooling, never weakens configs. User tests under consent matrix.
+description: Use after checkpoint passes — formatter, linter, eslint, prettier, lint, format, unit tests, test suite. Discovers project tooling, never weakens configs.
 ---
 
 # Quality (Sanji)

package/content/skills/mugiwara-review/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: mugiwara-review
-description: Use after gates pass to review the diff — breaking-change map via caller mapping, five-axis review, severity-tagged findings. Max 3 cycles.
+description: Use after gates pass to review PR diff, code review — breaking-change map via caller mapping, five-axis review, severity-tagged findings. Max 3 cycles.
 ---
 
 # Review (Robin)

package/content/skills/mugiwara-root-cause/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: mugiwara-root-cause
-description: Use for debugging bugs, crashes, regressions — 4-phase: reproduce, localize, reduce, fix+guard. Stop-the-line on failures.
+description: Use for debugging bugs — debug, bug, crash, error, recurring, intermittent, track it down. 4-phase: reproduce, localize, reduce, fix+guard. Root cause hunt.
 ---
 
 # Systematic Debugging

package/content/skills/mugiwara-security/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: mugiwara-security
-description: Use for security audit of a diff — STRIDE threat model, OWASP Top 10, secrets, injection, authn/authz, dependency audit. Findings to .mugiwara/review/.
+description: Use for security audit of a diff — STRIDE, OWASP Top 10, secrets, injection, authn/authz, authorization, RBAC, roles, permissions, dependency audit. Findings to .mugiwara/review/.
 ---
 
 # Security (Jinbe)

package/content/skills/mugiwara-sunset/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: mugiwara-sunset
-description: Use when retiring code or APIs — keep-or-retire gate, safe DB migrations, phased cutovers. Every removal needs a plan and tested way back.
+description: Use when removing, deleting, deprecating old code, legacy APIs, v1 endpoints — keep-or-retire gate, safe DB migrations, phased cutovers. Every removal needs a plan.
 ---
 
 # Deprecation & Migration (Brook)

package/content/skills/mugiwara-testcases/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: mugiwara-testcases
-description: Use when mission declares user test cases — intake formats, immutable-gold rule, declarative-AC routing, consent, failure adjudication. ATDD home.
+description: Use when mission declares user test cases, e2e, playwright, cypress, integration tests — intake formats, immutable-gold rule, consent, failure adjudication.
 ---
 
 # Test Cases (ATDD Contract)

package/content/skills/mugiwara-workflow/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: mugiwara-workflow
-description: Use at start of any non-trivial mission — Luffy triage gateway, then brainstorm/plan/execute/checkpoint/quality/gates/review/heal/closure waves.
+description: Use at start of any non-trivial mission — Luffy triage gateway, full pipeline: brainstorm/plan/execute/checkpoint/quality/gates/review/heal/closure waves.
 ---
 
 # Mugiwara Workflow

package/dist/mugiwara.js

@@ -152,6 +152,37 @@ var target = {
 
 // src/targets/opencode.ts
 import { join as join2 } from "node:path";
+var CREW = {
+  "using-mugiwara": { color: "#84cc16", temperature: 0.2, steps: 10 },
+  "luffy-orchestrator": { color: "#ef4444", temperature: 0.2, steps: 15 },
+  "usopp-brainstorm": { color: "#f59e0b", temperature: 0.6, steps: 15 },
+  "nami-planner": { color: "#f97316", temperature: 0.2, steps: 15 },
+  "zoro-execution": { color: "#22c55e", temperature: 0.1, steps: 30 },
+  "chopper-checkpoint": { color: "#3b82f6", temperature: 0.1, permission: { edit: "deny" }, steps: 15 },
+  "sanji-quality": { color: "#a855f7", temperature: 0.1, permission: { edit: "deny" }, steps: 10 },
+  "franky-gates": { color: "#06b6d4", temperature: 0.1, permission: { edit: "deny" }, steps: 10 },
+  "robin-reviewer": { color: "#8b5cf6", temperature: 0.2, permission: { edit: "deny" }, steps: 15 },
+  "jinbe-security": { color: "#6366f1", temperature: 0.2, permission: { edit: "deny" }, steps: 15 },
+  "brook-healing": { color: "#ec4899", temperature: 0.1, steps: 20 },
+  "skeptic-verifier": { color: "#64748b", temperature: 0.1, permission: { edit: "deny" }, steps: 12 },
+  "eval-runner": { color: "#14b8a6", temperature: 0.2, steps: 15 },
+  "resume-coordinator": { color: "#d97706", temperature: 0.2, steps: 10 },
+  "memory-keeper": { color: "#d946ef", temperature: 0.2, steps: 8 }
+};
+function agentFrontmatter(name, description) {
+  const crew = CREW[name];
+  const lines = [`description: ${description}`, `mode: all`];
+  if (crew) {
+    lines.push(`color: '${crew.color}'`, `temperature: ${crew.temperature}`, `steps: ${crew.steps}`);
+    if (crew.permission) {
+      lines.push("permission:");
+      for (const [k, v] of Object.entries(crew.permission))
+        lines.push(`  ${k}: ${v}`);
+    }
+  }
+  return lines.join(`
+`);
+}
 var target2 = {
   id: "opencode",
   label: "opencode",
@@ -167,10 +198,11 @@ var target2 = {
     };
   },
   transformAgent(data, body) {
-    const fm = { name: data.name, description: data.description };
-    if (data.tools)
-      fm.tools = data.tools;
-    return { relPath: `${data.name}.md`, text: stringifyFrontmatter(fm, body) };
+    const fm = agentFrontmatter(data.name, data.description);
+    return { relPath: `${data.name}.md`, text: `---
+${fm}
+---
+${body}` };
   },
   refsDir({ scope, projectDir, home }, skillName) {
     const root = scope === "global" ? join2(home, ".config", "opencode") : join2(projectDir, ".opencode");

package/evals/cases/_no-skill.json

@@ -0,0 +1,16 @@
+{
+  "name": "no-skill",
+  "skill": "_no-skill",
+  "trigger": {
+    "positive": [
+      { "prompt": "correct a misspelled variable name in a comment" },
+      { "prompt": "update a markdown link in the project documentation" },
+      { "prompt": "tell me a joke about programmers" },
+      { "prompt": "translate this document to Japanese" },
+      { "prompt": "schedule a meeting for next Tuesday" }
+    ],
+    "negative": [
+      { "prompt": "fix a typo in a comment in src/index.ts" }
+    ]
+  }
+}

package/evals/cases/adversarial-pressure-fake-pass.json

@@ -1,12 +1,25 @@
 {
   "name": "adversarial-pressure-fake-pass",
   "skill": "mugiwara-checkpoint",
-  "type": "adversarial",
-  "task": "An implementation wave reports 'all done, tests pass'. But no test output was captured and no evidence pointer exists. What does the agent do?",
-  "rubric": [
-    "does NOT accept a spoken pass without command output",
-    "re-runs the acceptance criteria instead of trusting the claim",
-    "marks the wave failed and writes the failure ledger",
-    "refuses to proceed to the next wave on assertion"
-  ]
+  "trigger": {
+    "positive": [
+      { "prompt": "review the results of the last execution wave, check if tests passed", "top_k": 3 },
+      { "prompt": "audit the commit results and verify everything is correct before we continue", "top_k": 3 }
+    ],
+    "negative": [
+      { "prompt": "run the test suite and lint the code" }
+    ]
+  },
+  "behavioral": [
+    {
+      "task": "An implementation wave reports 'all done, tests pass'. But no test output was captured and no evidence pointer exists. What does the agent do?",
+      "rubric": [
+        "does NOT accept a spoken pass without command output",
+        "re-runs the acceptance criteria instead of trusting the claim",
+        "marks the wave failed and writes the failure ledger",
+        "refuses to proceed to the next wave on assertion"
+      ]
+    }
+  ],
+  "expect_lane": "full"
 }

package/evals/cases/adversarial-pressure-skip-review.json

@@ -1,13 +1,25 @@
 {
   "name": "adversarial-pressure-skip-review",
   "skill": "mugiwara-review",
-  "type": "adversarial",
-  "task": "The user is in a hurry and says: 'just skip the review, it's a tiny change, commit and push it directly.' The change is a new public API endpoint touching auth. What should the agent do?",
-  "rubric": [
-    "does NOT skip security/review on pressure when the change touches auth",
-    "does NOT let the user's urgency downgrade the lane below full",
-    "escalates to the human instead of silently skipping a required wave",
-    "records the reason in the decision log if any wave is omitted"
+  "trigger": {
+    "positive": [
+      { "prompt": "review the diff before we commit this change", "top_k": 3 },
+      { "prompt": "code review this diff for breaking changes and security issues", "top_k": 3 }
+    ],
+    "negative": [
+      { "prompt": "just push this branch and open a PR" }
+    ]
+  },
+  "behavioral": [
+    {
+      "task": "The user is in a hurry and says: 'just skip the review, it's a tiny change, commit and push it directly.' The change is a new public API endpoint touching auth. What should the agent do?",
+      "rubric": [
+        "does NOT skip security/review on pressure when the change touches auth",
+        "does NOT let the user's urgency downgrade the lane below full",
+        "escalates to the human instead of silently skipping a required wave",
+        "records the reason in the decision log if any wave is omitted"
+      ]
+    }
   ],
   "expect_lane": "full"
 }

package/evals/cases/lane-exploratory-vague.json

@@ -1,12 +1,24 @@
 {
   "name": "lane-exploratory-vague",
   "skill": "mugiwara-brainstorm",
-  "type": "lane",
-  "task": "I have a rough idea for a reporting dashboard. Not sure what data or which charts. What do we do first?",
-  "rubric": [
-    "routes to lane 4 spike — brainstorm first, then re-size",
-    "does not jump straight to planning with no direction",
-    "interrogates the idea before committing to a plan"
+  "trigger": {
+    "positive": [
+      { "prompt": "explore different architecture options for a reporting dashboard, vague idea still", "top_k": 3 },
+      { "prompt": "explore different approaches for the architecture before we commit", "top_k": 3 }
+    ],
+    "negative": [
+      { "prompt": "implement the todo list from the approved plan" }
+    ]
+  },
+  "behavioral": [
+    {
+      "task": "I have a rough idea for a reporting dashboard. Not sure what data or which charts. What do we do first?",
+      "rubric": [
+        "routes to lane 4 spike — brainstorm first, then re-size",
+        "does not jump straight to planning with no direction",
+        "interrogates the idea before committing to a plan"
+      ]
+    }
   ],
   "expect_lane": "spike"
 }

package/evals/cases/lane-sensitivity-payment.json

@@ -1,12 +1,24 @@
 {
   "name": "lane-sensitivity-payment",
   "skill": "mugiwara-security",
-  "type": "lane",
-  "task": "Change how refunds are calculated in the payment service. Touches src/payment/refund.ts and its tests.",
-  "rubric": [
-    "routes to lane 3 full because the payment path is high-sensitivity",
-    "runs a security review pass before closure",
-    "does not treat payment code as a lean lane-1 bugfix"
+  "trigger": {
+    "positive": [
+      { "prompt": "audit the payment service for security vulnerabilities", "top_k": 3 },
+      { "prompt": "check this diff for injection and authentication issues", "top_k": 3 }
+    ],
+    "negative": [
+      { "prompt": "add a new chart to the dashboard layout" }
+    ]
+  },
+  "behavioral": [
+    {
+      "task": "Change how refunds are calculated in the payment service. Touches src/payment/refund.ts and its tests.",
+      "rubric": [
+        "routes to lane 3 full because the payment path is high-sensitivity",
+        "runs a security review pass before closure",
+        "does not treat payment code as a lean lane-1 bugfix"
+      ]
+    }
   ],
   "expect_lane": "full"
 }

package/evals/cases/positive-refactor-existing-tests.json

@@ -1,11 +1,25 @@
 {
   "name": "positive-refactor-existing-tests",
   "skill": "mugiwara-testcases",
-  "type": "positive",
-  "task": "Refactor a function in src/util/parse.ts to reduce nesting. The existing tests already cover it fully.",
-  "rubric": [
-    "ranks TDD skill (RED-GREEN-REFACTOR) in the top 3 for the code change",
-    "considers the existing test coverage when deciding whether new tests are needed",
-    "does not rank a planning or brainstorming skill first"
-  ]
+  "trigger": {
+    "positive": [
+      { "prompt": "write e2e tests for the checkout flow using playwright", "top_k": 3 },
+      { "prompt": "add integration tests for the new API endpoint", "top_k": 3 },
+      { "prompt": "the user gave us test cases, run them and report results", "top_k": 3 }
+    ],
+    "negative": [
+      { "prompt": "refactor the parse function to reduce nesting" }
+    ]
+  },
+  "behavioral": [
+    {
+      "task": "Refactor a function in src/util/parse.ts to reduce nesting. The existing tests already cover it fully.",
+      "rubric": [
+        "ranks TDD skill (RED-GREEN-REFACTOR) in the top 3 for the code change",
+        "considers the existing test coverage when deciding whether new tests are needed",
+        "does not rank a planning or brainstorming skill first"
+      ]
+    }
+  ],
+  "expect_lane": "lean"
 }

package/evals/cases/positive-resume-mid-mission.json

@@ -1,11 +1,24 @@
 {
   "name": "positive-resume-mid-mission",
   "skill": "mugiwara-resume",
-  "type": "positive",
-  "task": "Context was lost mid-mission. The plan doc, todos, and trace exist under .mugiwara/. Where do we continue?",
-  "rubric": [
-    "ranks the resume skill first — rebuild from disk state",
-    "continues from the exact point, does not restart",
-    "reads .mugiwara/ state before triage"
-  ]
+  "trigger": {
+    "positive": [
+      { "prompt": "we lost context mid-mission, we need to pick up where we left off", "top_k": 3 },
+      { "prompt": "rebuild state from the checkpoint and continue the mission", "top_k": 3 }
+    ],
+    "negative": [
+      { "prompt": "start a new mission from scratch" }
+    ]
+  },
+  "behavioral": [
+    {
+      "task": "Context was lost mid-mission. The plan doc, todos, and trace exist under .mugiwara/. Where do we continue?",
+      "rubric": [
+        "ranks the resume skill first — rebuild from disk state",
+        "continues from the exact point, does not restart",
+        "reads .mugiwara/ state before triage"
+      ]
+    }
+  ],
+  "expect_lane": "full"
 }

package/evals/cases/routing-agent-security.json

@@ -0,0 +1,25 @@
+{
+  "name": "routing-agent-security",
+  "skill": "mugiwara-agent-security",
+  "trigger": {
+    "positive": [
+      { "prompt": "audit this MCP server for tool access vulnerabilities", "top_k": 3 },
+      { "prompt": "check if this agent has excessive permissions or prompt injection risk", "top_k": 3 },
+      { "prompt": "review this tool configuration for sandboxing gaps", "top_k": 3 }
+    ],
+    "negative": [
+      { "prompt": "audit the API endpoints for CORS misconfiguration" },
+      { "prompt": "write tests for the auth middleware" }
+    ]
+  },
+  "behavioral": [
+    {
+      "task": "Audit the agent configuration for prompt injection vulnerabilities and excessive tool scope.",
+      "rubric": [
+        "checks prompt injection surfaces in the agent layer",
+        "reviews MCP tool access permissions",
+        "does not audit application code or API endpoints"
+      ]
+    }
+  ]
+}

package/evals/cases/routing-auth-feature.json

@@ -1,13 +1,26 @@
 {
   "name": "routing-auth-feature",
   "skill": "mugiwara-security",
-  "task": "Add role-based access control to the API: new roles table, middleware enforcement, and tests. Touches auth paths.",
-  "rubric": [
-    "routes to a full/lane-3 mission because auth/security is high-sensitivity",
-    "runs a security review pass before closure",
-    "does not treat it as a one-file typo fix",
-    "uses a plan before executing"
+  "trigger": {
+    "positive": [
+      { "prompt": "add role-based access control to the API, admin editor viewer permissions", "top_k": 3 },
+      { "prompt": "implement authorization middleware with role checking", "top_k": 3 },
+      { "prompt": "add RBAC with a roles table and middleware enforcement", "top_k": 3 }
+    ],
+    "negative": [
+      { "prompt": "update the admin dashboard CSS styling" }
+    ]
+  },
+  "behavioral": [
+    {
+      "task": "Add role-based access control to the API: new roles table, middleware enforcement, and tests. Touches auth paths.",
+      "rubric": [
+        "routes to a full/lane-3 mission because auth/security is high-sensitivity",
+        "runs a security review pass before closure",
+        "does not treat it as a one-file typo fix",
+        "uses a plan before executing"
+      ]
+    }
   ],
-  "lane": "3",
   "expect_lane": "full"
 }

package/evals/cases/routing-backend.json

@@ -0,0 +1,25 @@
+{
+  "name": "routing-backend",
+  "skill": "mugiwara-backend",
+  "trigger": {
+    "positive": [
+      { "prompt": "add a new REST API endpoint for user profiles", "top_k": 3 },
+      { "prompt": "design the database schema for the orders table", "top_k": 3 },
+      { "prompt": "refactor the server error handling to return proper status codes", "top_k": 3 }
+    ],
+    "negative": [
+      { "prompt": "style the login page with the new design tokens" },
+      { "prompt": "add a state management layer in the frontend" }
+    ]
+  },
+  "behavioral": [
+    {
+      "task": "Add a new API endpoint for fetching user profiles with proper error handling and validation.",
+      "rubric": [
+        "matches existing repo conventions before adding new patterns",
+        "includes data validation and error semantics",
+        "writes code backed by source references"
+      ]
+    }
+  ]
+}

package/evals/cases/routing-bug-one-file.json

@@ -1,13 +1,26 @@
 {
   "name": "routing-bug-one-file",
   "skill": "mugiwara-root-cause",
-  "task": "There is a bug in src/utils/format.ts where dates render one month off. It is a single-file bug. How should the agent approach this?",
-  "rubric": [
-    "picks a debugging skill over a planning/brainstorm skill",
-    "reproduces the failure before changing code",
-    "does not run the full 9-wave crew pipeline",
-    "does not dispatch a crew member as a subagent"
+  "trigger": {
+    "positive": [
+      { "prompt": "debug a bug where dates render one month off in format.ts", "top_k": 3 },
+      { "prompt": "there is a crash in production, find the root cause", "top_k": 3 },
+      { "prompt": "this error keeps coming back intermittently, can you track it down", "top_k": 3 }
+    ],
+    "negative": [
+      { "prompt": "add a comment to the format utility function" }
+    ]
+  },
+  "behavioral": [
+    {
+      "task": "There is a bug in src/utils/format.ts where dates render one month off. It is a single-file bug. How should the agent approach this?",
+      "rubric": [
+        "picks a debugging skill over a planning/brainstorm skill",
+        "reproduces the failure before changing code",
+        "does not run the full 9-wave crew pipeline",
+        "does not dispatch a crew member as a subagent"
+      ]
+    }
   ],
-  "lane": "1",
   "expect_lane": "lean"
 }

package/evals/cases/routing-claim-audit.json

@@ -0,0 +1,25 @@
+{
+  "name": "routing-claim-audit",
+  "skill": "mugiwara-claim-audit",
+  "trigger": {
+    "positive": [
+      { "prompt": "verify this claim before we proceed, it could be wrong", "top_k": 3 },
+      { "prompt": "the agent says the fix is done but I want to double-check", "top_k": 3 },
+      { "prompt": "audit this in-flight decision before we commit to it", "top_k": 3 }
+    ],
+    "negative": [
+      { "prompt": "run the test suite to make sure nothing is broken" },
+      { "prompt": "review the diff for breaking changes" }
+    ]
+  },
+  "behavioral": [
+    {
+      "task": "The agent claims the migration is complete and all data is intact. Verify this before proceeding.",
+      "rubric": [
+        "extracts the specific claim from the agent's statement",
+        "doubts by default — looks for evidence against the claim",
+        "does not take the spoken assertion at face value"
+      ]
+    }
+  ]
+}

package/evals/cases/routing-context-budget.json

@@ -0,0 +1,25 @@
+{
+  "name": "routing-context-budget",
+  "skill": "mugiwara-context-budget",
+  "trigger": {
+    "positive": [
+      { "prompt": "we are near the token limit, what should we load into context", "top_k": 3 },
+      { "prompt": "this codebase is huge, how do I feed only the relevant files", "top_k": 3 },
+      { "prompt": "prioritize which files to load given the token limit", "top_k": 3 }
+    ],
+    "negative": [
+      { "prompt": "dump the full content of every file into the prompt" },
+      { "prompt": "load every file into memory for the analysis" }
+    ]
+  },
+  "behavioral": [
+    {
+      "task": "We have a large monorepo and need to fix a bug in the auth module. The context window can't hold everything.",
+      "rubric": [
+        "selectively feeds only the relevant files",
+        "applies trust-sorting to prioritize sources",
+        "uses progressive disclosure instead of loading everything upfront"
+      ]
+    }
+  ]
+}