feat(mcp): add hivemind active task map

.changeset/hivemind-mcp-tool.md

@@ -0,0 +1,5 @@
+---
+"@cavemem/mcp-server": minor
+---
+
+Add a compact `hivemind` MCP tool that maps active proxy-runtime agent sessions to their current tasks.

.codex/skills/reviewer/SKILL.md

@@ -0,0 +1,52 @@
+---
+name: reviewer
+description: Use when reviewing active multi-agent work, PRs, or handoffs where you need Cavemem's hivemind tool to map branch and task ownership before reading full session memory.
+---
+
+# Reviewer
+
+Use this skill for review-shaped tasks where ownership matters:
+
+- PR review on an active agent branch
+- Handoff review
+- "who owns this task?"
+- conflict triage across worktrees
+- stale lane vs live lane checks before touching code
+
+## Fast path
+
+1. Call `hivemind` with the target `repo_root` (or `repo_roots`).
+2. Pick the smallest relevant session set by exact `branch`, `task`, `agent`, or `worktree_path`.
+3. Only after you know which session matters:
+   - use `search` for cross-session topic lookup
+   - or `list_sessions` -> `timeline` when you already know the session
+4. Fetch full bodies with `get_observations` only for the exact evidence you need.
+5. Review the code or diff with findings first: bugs, regressions, missing tests, or ownership conflicts.
+
+## Hivemind rules
+
+- Treat `activity=working` or `activity=thinking` as a live lane.
+- Treat `source=worktree-lock` without a matching active session as fallback telemetry, not proof.
+- Prefer the freshest exact `branch` match over loose task-name similarity.
+- Do not fetch observation bodies for every session "just in case". `hivemind` exists to stop that waste.
+
+## Review checklist
+
+- Confirm which branch or worktree owns the task.
+- Check whether the lane is live, stalled, or dead.
+- Pull only the memory evidence needed for the review claim.
+- Review the actual diff or files after ownership is clear.
+- Report findings first, highest risk first.
+
+## Output contract
+
+- Findings first.
+- Cite the `branch`, `worktree_path`, or session evidence you used.
+- Call out stale telemetry or ownership collisions explicitly.
+- If no findings, say so and mention residual risk or missing verification.
+
+## Example prompts
+
+- `Review PR #2 with hivemind first so you know which branch still owns the lane.`
+- `Check whether this handoff is stale or still live before I edit the worktree.`
+- `Find who owns the runtime bug, then review only that lane's evidence.`

.codex/skills/reviewer/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "Cavemem Reviewer"
+  short_description: "Review active lanes with hivemind-first context"
+  default_prompt: "Review this active repo or PR by using Cavemem hivemind first, then fetch only the minimum memory and code context needed for findings."

apps/cli/vitest.config.ts

@@ -1,8 +1,12 @@
 import { readFileSync } from 'node:fs';
-import { defineConfig } from 'vitest/config';
+import { defineConfig, mergeConfig } from 'vitest/config';
+import rootConfig from '../../vitest.config.js';
 
 const { version } = JSON.parse(readFileSync('./package.json', 'utf8')) as { version: string };
 
-export default defineConfig({
-  define: { __CAVEMEM_VERSION__: JSON.stringify(version) },
-});
+export default mergeConfig(
+  rootConfig,
+  defineConfig({
+    define: { __CAVEMEM_VERSION__: JSON.stringify(version) },
+  }),
+);