Isolate judge MCP profiles

agents/api-security-reviewer.md

@@ -1,7 +1,8 @@
 ---
 name: api-security-reviewer
+mcpServers: [codebase, context7, autonomous_bugs]
 description: "Use this agent to review backend/API code for security vulnerabilities. Checks authentication, JWT handling, input validation, output security, and access control.\\n\\n<example>\\nContext: User has added new API endpoints or authentication logic.\\nuser: \"I've added the new user registration endpoint\"\\nassistant: \"Let me invoke the api-security-reviewer agent to check for security issues in your new endpoint.\"\\n<commentary>\\nNew API endpoints should be reviewed for authentication, input validation, and secure response handling.\\n</commentary>\\n</example>\\n\\n<example>\\nContext: User has modified JWT or session handling.\\nuser: \"Updated the token refresh logic\"\\nassistant: \"I'll run the api-security-reviewer agent to verify the JWT implementation follows security best practices.\"\\n<commentary>\\nJWT changes require verification of algorithm, expiry, and secret handling.\\n</commentary>\\n</example>"
-tools: Read, Grep, Glob, Bash
+tools: Read, Grep, Glob, Bash, mcp__codebase, mcp__context7, mcp__autonomous_bugs
 model: haiku
 color: red
 ---

agents/backend-performance-reviewer.md

@@ -1,7 +1,8 @@
 ---
 name: backend-performance-reviewer
+mcpServers: [codebase, context7, autonomous_bugs]
 description: "Use this agent to review backend code for performance issues. Checks database queries, caching strategies, async patterns, and API response optimization.\\n\\n<example>\\nContext: User has added database queries or data fetching logic.\\nuser: \"Added the query to fetch all user tasks\"\\nassistant: \"Let me invoke the backend-performance-reviewer agent to check for N+1 queries and pagination issues.\"\\n<commentary>\\nDatabase queries should be reviewed for efficiency, proper indexing, and avoiding N+1 patterns.\\n</commentary>\\n</example>\\n\\n<example>\\nContext: User has implemented caching or heavy data processing.\\nuser: \"Implemented caching for the dashboard data\"\\nassistant: \"I'll run the backend-performance-reviewer agent to verify cache invalidation and TTL strategies.\"\\n<commentary>\\nCaching implementations need review for proper invalidation and memory considerations.\\n</commentary>\\n</example>"
-tools: Read, Grep, Glob, Bash
+tools: Read, Grep, Glob, Bash, mcp__codebase, mcp__context7, mcp__autonomous_bugs
 model: haiku
 color: orange
 ---

agents/commit-guard.md

@@ -1,7 +1,8 @@
 ---
 name: commit-guard
+mcpServers: [codebase, context7, autonomous_bugs]
 description: "Use this agent before committing code to guard against unintentional duplication using semantic search.\\n\\n<example>\\nContext: User is about to commit staged changes.\\nuser: \"I'm ready to commit these changes\"\\nassistant: \"Let me invoke the commit-guard agent to check for duplicates before you commit.\"\\n<commentary>\\ncommit-guard runs semantic duplicate detection against the search index and checks DRY rules per file.\\n</commentary>\\n</example>\\n\\n<example>\\nContext: User has added new utility functions or components.\\nuser: \"Added a new helper function for date formatting\"\\nassistant: \"I'll run the commit-guard agent to verify this doesn't duplicate an existing utility.\"\\n<commentary>\\nNew utilities should be checked against the indexed codebase to prevent duplication.\\n</commentary>\\n</example>"
-tools: Read, Grep, Glob, Bash, mcp__codebase__searchCode
+tools: Read, Grep, Glob, Bash, mcp__codebase, mcp__context7, mcp__autonomous_bugs, mcp__codebase__searchCode
 model: haiku
 color: blue
 ---

agents/conventions-reviewer.md

@@ -1,7 +1,8 @@
 ---
 name: conventions-reviewer
+mcpServers: [codebase, context7, autonomous_bugs]
 description: "Use this agent to check a diff against the governing CLAUDE.md files of the repo it's installed in. Flags a violation only when it can quote both the exact rule and the exact offending line; otherwise stays silent. No style opinions.\\n\\n<example>\\nContext: A CLAUDE.md rule says never hand-edit generated files.\\nuser: \"Updated the generated icon exports directly\"\\nassistant: \"I'll run the conventions-reviewer agent to check whether that edit violates the repo's own generated-file rule.\"\\n<commentary>\\nA written CLAUDE.md rule with an unhedged directive and a concrete offending line is exactly what this reviewer exists to catch.\\n</commentary>\\n</example>\\n\\n<example>\\nContext: A nested package has its own CLAUDE.md scoping a rule to that package only.\\nuser: \"Added a new file under packages/api\"\\nassistant: \"Let me invoke the conventions-reviewer agent — packages/api's own CLAUDE.md may govern this file, on top of the repo root's.\"\\n<commentary>\\nScoping matters: a rule in one package's CLAUDE.md never governs a sibling package's files.\\n</commentary>\\n</example>"
-tools: Read, Grep, Glob
+tools: Read, Grep, Glob, mcp__codebase, mcp__context7, mcp__autonomous_bugs
 model: haiku
 color: cyan
 ---

agents/correctness-reviewer.md

@@ -1,7 +1,8 @@
 ---
 name: correctness-reviewer
+mcpServers: [codebase, context7, autonomous_bugs]
 description: "Use this agent to review a finished diff for correctness bugs: concurrency/race conditions, state-machine dead states, writer/reader contract mismatches across modules, broadcast/dedup errors, discarded return values, and classifier/parsing edge cases.\\n\\n<example>\\nContext: User has implemented retry/recovery logic that writes task statuses.\\nuser: \"The task retry flow is done\"\\nassistant: \"Let me invoke the correctness-reviewer agent to trace every status write to its readers and walk the concurrent interleavings.\"\\n<commentary>\\nStatus writes need CAS guards and every consumer (pollers, filters, queries) must still select the written state.\\n</commentary>\\n</example>\\n\\n<example>\\nContext: User added an event that is broadcast to multiple listeners.\\nuser: \"chat-ready now fires on resume\"\\nassistant: \"I'll run the correctness-reviewer agent to enumerate the listeners and check for duplicate handling.\"\\n<commentary>\\nA broadcast consumed by N listeners without targeting/dedup executes its effect N times.\\n</commentary>\\n</example>"
-tools: Read, Grep, Glob, Bash
+tools: Read, Grep, Glob, Bash, mcp__codebase, mcp__context7, mcp__autonomous_bugs
 model: sonnet
 color: orange
 ---

agents/feature-completeness-reviewer.md

@@ -1,5 +1,6 @@
 ---
 name: feature-completeness-reviewer
+mcpServers: [codebase, context7, autonomous_bugs]
 model: opus
 description: Use this agent proactively after any significant code or feature implementation to identify missing components, overlooked dependencies, and gaps in the solution. Specifically invoke this agent when:\\n\\n<example>\\nContext: The user has just implemented a new feature for real-time chat functionality.\\nuser: \"I've just finished implementing the core chat messaging system\"\\nassistant: \"Great work on the messaging system! Let me use the Task tool to launch the feature-completeness-reviewer agent to analyze what supporting components might be needed.\"\\n<commentary>\\nSince a significant feature was implemented, proactively use the feature-completeness-reviewer agent to identify missing pieces like debugging tools, monitoring capabilities, or infrastructure requirements.\\n</commentary>\\n</example>\\n\\n<example>\\nContext: The user has completed a major refactoring of the authentication system.\\nuser: \"The authentication refactor is done, all tests passing\"\\nassistant: \"Excellent! Now let me invoke the feature-completeness-reviewer agent to examine if we've covered all the necessary touchpoints.\"\\n<commentary>\\nAfter completing authentication work, use the feature-completeness-reviewer agent to check for missing elements like session management, token refresh mechanisms, or user experience flows.\\n</commentary>\\n</example>\\n\\n<example>\\nContext: A new API endpoint has been added to the system.\\nuser: \"I've added the new /api/status endpoint\"\\nassistant: \"Perfect! Let me use the feature-completeness-reviewer agent to assess the broader implications and any missing pieces.\"\\n<commentary>\\nWhen new APIs are created, proactively launch the feature-completeness-reviewer agent to identify missing documentation, error handling, rate limiting, or integration points.\\n</commentary>\\n</example>
 ---

agents/feature-critique.md

@@ -1,5 +1,6 @@
 ---
 name: feature-critique
+mcpServers: [codebase, context7, autonomous_bugs]
 model: opus
 description: Pre-implementation critique specialist. Analyzes feature proposals, architectural decisions, and technical approaches BEFORE work begins. Evaluates feasibility, UX implications, security concerns, codebase conflicts, data flow correctness, runtime behavior across user configurations, and missing considerations using evidence-based research. Invoke when a feature proposal, technical approach, or architectural decision needs critical evaluation before implementation starts.
 ---

agents/fix-upstream-reviewer.md

@@ -1,5 +1,6 @@
 ---
 name: fix-upstream-reviewer
+mcpServers: [codebase, context7, autonomous_bugs]
 model: opus
 description: Lead-engineer code reviewer that identifies fallbacks, workarounds, and patches that mask upstream problems. Use proactively when reviewing code that contains try/catch fallbacks, redundant API calls, defensive null-checks, or multi-path resolution logic. Answers two questions - can we simplify this code, and should we fix upstream (closer to the source of the problem) instead of patching downstream? Here "upstream" means the direction the data flows FROM, not any particular repo.
 ---

agents/frontend-accessibility-reviewer.md

@@ -1,7 +1,8 @@
 ---
 name: frontend-accessibility-reviewer
+mcpServers: [codebase, context7, autonomous_bugs]
 description: "Use this agent to review frontend code for accessibility (WCAG AA) issues. Checks semantic HTML, keyboard navigation, ARIA usage, color contrast, form labels, focus management, and motion safety. Advisory only.\\n\\n<example>\\nContext: User has added a new interactive component.\\nuser: \"Added the new dropdown menu\"\\nassistant: \"Let me invoke the frontend-accessibility-reviewer agent to check keyboard navigation, focus management, and ARIA usage.\"\\n<commentary>\\nNew interactive widgets should be reviewed for keyboard access and correct ARIA roles.\\n</commentary>\\n</example>\\n\\n<example>\\nContext: User has changed colors or added text on a colored background.\\nuser: \"Updated the badge styles\"\\nassistant: \"I'll run the frontend-accessibility-reviewer agent to verify contrast ratios meet WCAG AA.\"\\n<commentary>\\nColor changes need a contrast check (4.5:1 normal text, 3:1 large/non-text).\\n</commentary>\\n</example>"
-tools: Read, Grep, Glob, Bash
+tools: Read, Grep, Glob, Bash, mcp__codebase, mcp__context7, mcp__autonomous_bugs
 model: opus
 color: blue
 ---

agents/frontend-performance-reviewer.md

@@ -1,7 +1,8 @@
 ---
 name: frontend-performance-reviewer
+mcpServers: [codebase, context7, autonomous_bugs]
 description: "Use this agent to review frontend code for performance issues. Checks bundle size, image optimization, CSS efficiency, and React rendering patterns.\\n\\n<example>\\nContext: User has added new React components or modified rendering logic.\\nuser: \"Added the new dashboard widgets\"\\nassistant: \"Let me invoke the frontend-performance-reviewer agent to check for unnecessary re-renders and bundle size impact.\"\\n<commentary>\\nNew components should be reviewed for React.memo usage, proper hook dependencies, and lazy loading opportunities.\\n</commentary>\\n</example>\\n\\n<example>\\nContext: User has added images or modified CSS.\\nuser: \"Added the product images to the catalog page\"\\nassistant: \"I'll run the frontend-performance-reviewer agent to verify image optimization and lazy loading.\"\\n<commentary>\\nImages need review for proper formats, dimensions, and lazy loading implementation.\\n</commentary>\\n</example>"
-tools: Read, Grep, Glob, Bash
+tools: Read, Grep, Glob, Bash, mcp__codebase, mcp__context7, mcp__autonomous_bugs
 model: haiku
 color: yellow
 ---

agents/frontend-security-reviewer.md

@@ -1,7 +1,8 @@
 ---
 name: frontend-security-reviewer
+mcpServers: [codebase, context7, autonomous_bugs]
 description: "Use this agent to review frontend code for security vulnerabilities. Checks XSS prevention, CSRF protection, token storage, and input validation.\\n\\n<example>\\nContext: User has added form handling or user input processing.\\nuser: \"Added the comment submission form\"\\nassistant: \"Let me invoke the frontend-security-reviewer agent to check for XSS vulnerabilities and input sanitization.\"\\n<commentary>\\nUser input handling should be reviewed for proper sanitization and XSS prevention.\\n</commentary>\\n</example>\\n\\n<example>\\nContext: User has modified authentication or token handling.\\nuser: \"Updated the login flow to store the session\"\\nassistant: \"I'll run the frontend-security-reviewer agent to verify tokens aren't stored in localStorage and are properly secured.\"\\n<commentary>\\nAuthentication changes need review for secure token storage and CSRF protection.\\n</commentary>\\n</example>"
-tools: Read, Grep, Glob, Bash
+tools: Read, Grep, Glob, Bash, mcp__codebase, mcp__context7, mcp__autonomous_bugs
 model: haiku
 color: red
 ---

agents/prior-art.md

@@ -1,5 +1,6 @@
 ---
 name: prior-art
+mcpServers: [codebase, context7, autonomous_bugs]
 model: opus
 description: Step-0 problem validation BEFORE any plan exists. Given a problem statement, researches whether the problem is already solved (locally cloned reference checkouts, upstream fixes, other consumers of the same dependency, the web), whether it is a red herring, and whether the problem's frame itself should exist. Returns a cited SOLVED_ELSEWHERE / DISSOLVE_FRAME / GENUINE_NEW_WORK / INSUFFICIENT_EVIDENCE verdict with per-leg availability attestation. Invoke when a task is problem-shaped — a bug or pain attributed to a dependency, a missing capability, a limit to work around — before options or plans are drafted.
 ---

cli/lib/install/agent-assets/agent-assets.mts

@@ -5,6 +5,7 @@ import { type AgentAssetKind, type AgentProvider, isAgentProvider } from './agen
 const FRONTMATTER_RE = /^(?:\uFEFF)?---[ \t]*\r?\n([\s\S]*?)^---[ \t]*(?:\r?\n|$)/m;
 const FRONTMATTER_KEY_RE = /^[A-Za-z_][A-Za-z0-9_-]*$/;
 const AGENT_NAME_RE = /^[A-Za-z0-9][A-Za-z0-9_-]*$/;
+const MCP_SERVER_NAME_RE = /^[A-Za-z0-9][A-Za-z0-9_-]*$/;
 const NUMBER_RE = /^[+-]?(?:\d+(?:\.\d*)?|\.\d+)(?:[eE][+-]?\d+)?$/;
 const PLAIN_COMMENT_RE = /\s+#/;
 const NULL_RE = /^(?:null)$/i;
@@ -144,6 +145,20 @@ function parseFrontmatterScalar(raw: string, key: string): unknown {
   return parsePlainScalar(raw);
 }
 
+function parseMcpServers(raw: string): string[] {
+  if (!raw.startsWith('[') || !raw.endsWith(']'))
+    throw new Error('Agent frontmatter field "mcpServers" must be an inline string list');
+  const names = raw
+    .slice(1, -1)
+    .split(',')
+    .map((name) => name.trim());
+  if (names.length === 0 || names.some((name) => !MCP_SERVER_NAME_RE.test(name)))
+    throw new Error('Agent frontmatter field "mcpServers" contains an invalid server name');
+  if (new Set(names).size !== names.length)
+    throw new Error('Agent frontmatter field "mcpServers" contains a duplicate server name');
+  return names;
+}
+
 function requiredString(fields: Map<string, unknown>, key: string): string {
   const value = fields.get(key);
   if (typeof value !== 'string' || !value.trim())
@@ -171,7 +186,8 @@ function parseAgentFrontmatter(markdown: string): AgentFrontmatter {
       throw new Error(`Malformed agent frontmatter line ${index + 2}: ${line}`);
     }
     if (fields.has(key)) throw new Error(`Duplicate agent frontmatter field: ${key}`);
-    fields.set(key, parseFrontmatterScalar(line.slice(separator + 1).trim(), key));
+    const raw = line.slice(separator + 1).trim();
+    fields.set(key, key === 'mcpServers' ? parseMcpServers(raw) : parseFrontmatterScalar(raw, key));
   }
 
   const body = markdown.slice(match[0].length);

cli/lib/install/agent-assets/agent-assets.test.mts

@@ -42,6 +42,7 @@ describe('agent asset projections', () => {
       'name: quote-reviewer',
       String.raw`description: "Quotes: \"yes\"; path C:\\tmp; literal \\n marker" # metadata comment`,
       'tools: Read, Grep, Bash',
+      'mcpServers: [codebase, context7, autonomous_bugs]',
       'model: opus',
       'color: blue',
       '---',
@@ -64,6 +65,7 @@ Keep C:\tmp and triple quotes """ intact.
 `,
     });
     expect(toml).not.toContain('tools =');
+    expect(toml).not.toContain('mcpServers =');
     expect(toml).not.toContain('model =');
     expect(toml).not.toContain('color =');
 
@@ -104,6 +106,14 @@ Keep C:\tmp and triple quotes """ intact.
     ['tagged indented block scalar', '---\nname: reviewer\ndescription: !!str |2\n---\nbody'],
     ['tagged anchor', '---\nname: reviewer\ndescription: !!str &label true\n---\nbody'],
     ['collection description', '---\nname: reviewer\ndescription: [not, a, string]\n---\nbody'],
+    [
+      'block MCP server list',
+      '---\nname: reviewer\ndescription: valid\nmcpServers:\n  - codebase\n---\nbody',
+    ],
+    [
+      'invalid MCP server name',
+      '---\nname: reviewer\ndescription: valid\nmcpServers: [codebase, ../bad]\n---\nbody',
+    ],
     ['empty body', '---\nname: reviewer\ndescription: valid\n---\n'],
     ['malformed field', '---\nname reviewer\ndescription: valid\n---\nbody'],
     ['malformed quoted scalar', '---\nname: reviewer\ndescription: "bad\\q"\n---\nbody'],
@@ -134,4 +144,21 @@ Keep C:\tmp and triple quotes """ intact.
       ]);
     }
   });
+
+  it('gives every bundled Claude agent only the shared MCP baseline', () => {
+    const agentsDir = join(packageDir(), 'agents');
+    const files = readdirSync(agentsDir).filter((name) => name.endsWith('.md'));
+
+    for (const file of files) {
+      const markdown = readFileSync(join(agentsDir, file), 'utf8');
+      expect(markdown, file).toMatch(/^mcpServers: \[codebase, context7, autonomous_bugs\]$/m);
+      const tools = markdown.match(/^tools: (.+)$/m)?.[1];
+      if (tools) {
+        const toolNames = tools.split(',').map((tool) => tool.trim());
+        expect(toolNames, file).toEqual(
+          expect.arrayContaining(['mcp__codebase', 'mcp__context7', 'mcp__autonomous_bugs']),
+        );
+      }
+    }
+  });
 });

dist/agents/api-security-reviewer.md

@@ -1,7 +1,8 @@
 ---
 name: api-security-reviewer
+mcpServers: [codebase, context7, autonomous_bugs]
 description: "Use this agent to review backend/API code for security vulnerabilities. Checks authentication, JWT handling, input validation, output security, and access control.\\n\\n<example>\\nContext: User has added new API endpoints or authentication logic.\\nuser: \"I've added the new user registration endpoint\"\\nassistant: \"Let me invoke the api-security-reviewer agent to check for security issues in your new endpoint.\"\\n<commentary>\\nNew API endpoints should be reviewed for authentication, input validation, and secure response handling.\\n</commentary>\\n</example>\\n\\n<example>\\nContext: User has modified JWT or session handling.\\nuser: \"Updated the token refresh logic\"\\nassistant: \"I'll run the api-security-reviewer agent to verify the JWT implementation follows security best practices.\"\\n<commentary>\\nJWT changes require verification of algorithm, expiry, and secret handling.\\n</commentary>\\n</example>"
-tools: Read, Grep, Glob, Bash
+tools: Read, Grep, Glob, Bash, mcp__codebase, mcp__context7, mcp__autonomous_bugs
 model: haiku
 color: red
 ---

dist/agents/backend-performance-reviewer.md

@@ -1,7 +1,8 @@
 ---
 name: backend-performance-reviewer
+mcpServers: [codebase, context7, autonomous_bugs]
 description: "Use this agent to review backend code for performance issues. Checks database queries, caching strategies, async patterns, and API response optimization.\\n\\n<example>\\nContext: User has added database queries or data fetching logic.\\nuser: \"Added the query to fetch all user tasks\"\\nassistant: \"Let me invoke the backend-performance-reviewer agent to check for N+1 queries and pagination issues.\"\\n<commentary>\\nDatabase queries should be reviewed for efficiency, proper indexing, and avoiding N+1 patterns.\\n</commentary>\\n</example>\\n\\n<example>\\nContext: User has implemented caching or heavy data processing.\\nuser: \"Implemented caching for the dashboard data\"\\nassistant: \"I'll run the backend-performance-reviewer agent to verify cache invalidation and TTL strategies.\"\\n<commentary>\\nCaching implementations need review for proper invalidation and memory considerations.\\n</commentary>\\n</example>"
-tools: Read, Grep, Glob, Bash
+tools: Read, Grep, Glob, Bash, mcp__codebase, mcp__context7, mcp__autonomous_bugs
 model: haiku
 color: orange
 ---