milens 0.6.9 → 0.7.1

package/adapters/README.md

@@ -1,107 +1,144 @@
-# Milens Adapter Packs
-
-Ready-to-use configuration files for connecting milens MCP server to popular AI coding harnesses.
-
-## What is milens?
-
-Milens is a code intelligence MCP server that provides deep symbol search, dependency graphs, blast-radius analysis, and pre-commit change detection. These adapter packs wire milens into your AI coding tool so it can answer questions like "what calls this function?" and "what breaks if I change this?"
-
-## Available Adapters
-
-| Harness | Directory | What's included |
-|---|---|---|
-| **Claude Code** | `claude-code/` | `.claude/mcp.json` + `CLAUDE.md` |
-| **OpenCode** | `opencode/` | `.opencode/config.json` + `AGENTS.md` |
-| **Codex** | `codex/` | `.codex/codex.md` |
-| **Cursor** | `cursor/` | `.cursorrules` |
-| **GitHub Copilot** | `copilot/` | `.github/copilot-instructions.md` |
-| **Gemini** | `gemini/` | `.gemini/context.md` |
-| **Zed** | `zed/` | `.zed/settings.json` |
-
-## Installation
-
-### Claude Code
-
-```bash
-cp -r adapters/claude-code/.claude .claude/
-cp adapters/claude-code/CLAUDE.md CLAUDE.md
-```
-
-### OpenCode
-
-```bash
-cp -r adapters/opencode/.opencode .opencode/
-cp adapters/opencode/AGENTS.md AGENTS.md
-```
-
-### Codex
-
-```bash
-cp -r adapters/codex/.codex .codex/
-```
-
-### Cursor
-
-```bash
-cp adapters/cursor/.cursorrules .cursorrules
-```
-
-### GitHub Copilot
-
-```bash
-cp -r adapters/copilot/.github .github/
-```
-
-### Gemini
-
-```bash
-cp -r adapters/gemini/.gemini .gemini/
-```
-
-### Zed
-
-```bash
-cp -r adapters/zed/.zed .zed/
-```
-
-## Verification
-
-After installation, verify milens is connected:
-
-1. Start your AI harness
-2. Look for the milens MCP server in the tools list
-3. Run `mcp_milens_status` to confirm the index is healthy
-
-## Index Maintenance
-
-After significant code changes, refresh the index:
-
-```bash
-npx milens analyze -p . --force
-```
-
-## All Tools Available
-
-| Tool | Purpose |
-|---|---|
-| `mcp_milens_overview` | Combined context + impact + grep |
-| `mcp_milens_edit_check` | Pre-edit safety check |
-| `mcp_milens_impact` | Blast radius analysis |
-| `mcp_milens_context` | 360° caller/callee view |
-| `mcp_milens_query` | Symbol definitions by name |
-| `mcp_milens_grep` | Full-text search all files |
-| `mcp_milens_detect_changes` | Pre-commit change detection |
-| `mcp_milens_explain_relationship` | Shortest path between symbols |
-| `mcp_milens_get_file_symbols` | All symbols in a file |
-| `mcp_milens_get_type_hierarchy` | Class inheritance tree |
-| `mcp_milens_find_dead_code` | Unused exported symbols |
-| `mcp_milens_trace` | Execution flow tracing |
-| `mcp_milens_status` | Index health check |
-
-## Safety Rules (applies to all harnesses)
-
-- Never edit a symbol without `mcp_milens_edit_check` first
-- Never delete/rename without `mcp_milens_grep` + `mcp_milens_impact`
-- Never commit without `mcp_milens_detect_changes`
-- Use `mcp_milens_grep` for phrases/UI text, `mcp_milens_query` for code symbols
-- All milens calls require `repo` set to the absolute workspace root
+# Milens Adapter Packs
+
+Ready-to-use configuration files for connecting milens MCP server to popular AI coding harnesses.
+
+## What is milens?
+
+Milens is a code intelligence MCP server that provides deep symbol search, dependency graphs, blast-radius analysis, and pre-commit change detection. These adapter packs wire milens into your AI coding tool so it can answer questions like "what calls this function?" and "what breaks if I change this?"
+
+## Available Adapters
+
+| Harness | Directory | What's included |
+|---|---|---|
+| **Claude Code** | `claude-code/` | `.claude/mcp.json` + `CLAUDE.md` (+ optional hooks snippet) |
+| **OpenCode** | `opencode/` | `.opencode/config.json` + `AGENTS.md` |
+| **Codex** | `codex/` | `.codex/config.toml` + `.codex/codex.md` |
+| **Cursor** | `cursor/` | `.cursor/mcp.json` + `.cursorrules` |
+| **GitHub Copilot** | `copilot/` | `.vscode/mcp.json` + `.github/copilot-instructions.md` |
+| **Gemini** | `gemini/` | `.gemini/settings.json` + `.gemini/context.md` |
+| **Zed** | `zed/` | `.zed/settings.json` |
+
+## Installation
+
+### Claude Code
+
+Register milens as a **project-scoped MCP server** — this is the one method that works identically across the Claude Code CLI, the VS Code extension, and the desktop app, because it relies on nothing but a plain `.mcp.json` file at your project root (no plugin marketplace involved).
+
+```bash
+cp -r adapters/claude-code/.claude .claude/
+cp adapters/claude-code/CLAUDE.md CLAUDE.md
+claude mcp add milens -- milens serve -p .
+```
+
+That command writes a `.mcp.json` at your project root equivalent to:
+
+```json
+{
+  "mcpServers": {
+    "milens": {
+      "command": "milens",
+      "args": ["serve", "-p", "."]
+    }
+  }
+}
+```
+
+`-p .` is resolved against the process's working directory, which Claude Code always sets to your project root when it spawns the server — so this is the most portable form and needs no environment-variable substitution.
+
+> **Prerequisite:** `npm i -g milens`
+
+### Claude Code — via `/plugin install` (marketplace)
+
+```bash
+/plugin marketplace add fuze210699/milens
+/plugin install milens
+```
+
+This reads `.claude-plugin/marketplace.json` at the repo root, which points at `adapters/claude-code` as the plugin. That plugin ships its own `.mcp.json` with **no `-p` flag** — `milens serve` falls back to the `CLAUDE_PROJECT_DIR` environment variable, which Claude Code injects into every MCP server subprocess it spawns (project-scoped or plugin-scoped).
+
+> **History:** we previously shipped a plugin manifest that passed `${CLAUDE_PLUGIN_ROOT}` as `-p`. That variable resolves to the *plugin's own install directory* (e.g. `~/.claude/plugins/cache/...`), not your project — so it silently indexed the wrong codebase. We removed it in favor of the manual `.mcp.json` method above.
+>
+> The current plugin fixes this by relying on `CLAUDE_PROJECT_DIR` (env var, not a `${...}` template substitution) read directly by the CLI — `resolve(opts.path ?? process.env.CLAUDE_PROJECT_DIR ?? '.')` in `src/cli.ts`. Claude Code's docs don't specify a minimum version for `CLAUDE_PROJECT_DIR` injection into plugin-scoped servers, so **verify with `mcp_milens_status` after install** — if the repo path looks wrong, fall back to the manual `.mcp.json` method above, which is unaffected by any of this and always works.
+
+**Optional: session-start / pre-compact hooks.** Merge the `hooks` key from `adapters/claude-code/.claude/settings.json.hooks-snippet.json` into your project's `.claude/settings.json` (create the file if it doesn't exist yet). Not copied automatically since `.claude/settings.json` commonly holds other project settings you don't want overwritten.
+
+### OpenCode
+
+```bash
+cp -r adapters/opencode/.opencode .opencode/
+cp adapters/opencode/AGENTS.md AGENTS.md
+```
+
+### Codex
+
+```bash
+cp -r adapters/codex/.codex .codex/
+```
+
+### Cursor
+
+```bash
+cp -r adapters/cursor/.cursor .cursor/
+cp adapters/cursor/.cursorrules .cursorrules
+```
+
+### GitHub Copilot
+
+```bash
+cp -r adapters/copilot/.vscode .vscode/
+cp -r adapters/copilot/.github .github/
+```
+
+### Gemini
+
+```bash
+cp -r adapters/gemini/.gemini .gemini/
+```
+
+### Zed
+
+```bash
+cp -r adapters/zed/.zed .zed/
+```
+
+## Verification
+
+After installation, verify milens is connected:
+
+1. Start your AI harness
+2. Look for the milens MCP server in the tools list
+3. Run `mcp_milens_status` to confirm the index is healthy
+
+## Index Maintenance
+
+After significant code changes, refresh the index:
+
+```bash
+npx milens analyze -p . --force
+```
+
+## All Tools Available
+
+| Tool | Purpose |
+|---|---|
+| `mcp_milens_overview` | Combined context + impact + grep |
+| `mcp_milens_edit_check` | Pre-edit safety check |
+| `mcp_milens_impact` | Blast radius analysis |
+| `mcp_milens_context` | 360° caller/callee view |
+| `mcp_milens_query` | Symbol definitions by name |
+| `mcp_milens_grep` | Full-text search all files |
+| `mcp_milens_detect_changes` | Pre-commit change detection |
+| `mcp_milens_explain_relationship` | Shortest path between symbols |
+| `mcp_milens_get_file_symbols` | All symbols in a file |
+| `mcp_milens_get_type_hierarchy` | Class inheritance tree |
+| `mcp_milens_find_dead_code` | Unused exported symbols |
+| `mcp_milens_trace` | Execution flow tracing |
+| `mcp_milens_status` | Index health check |
+
+## Safety Rules (applies to all harnesses)
+
+- Never edit a symbol without `mcp_milens_edit_check` first
+- Never delete/rename without `mcp_milens_grep` + `mcp_milens_impact`
+- Never commit without `mcp_milens_detect_changes`
+- Use `mcp_milens_grep` for phrases/UI text, `mcp_milens_query` for code symbols
+- All milens calls require `repo` set to the absolute workspace root

package/adapters/claude-code/.claude/mcp.json

@@ -1,9 +1,9 @@
-{
-  "mcpServers": {
-    "milens": {
-      "command": "milens",
-      "args": ["serve", "-p", "."],
-      "env": { "MILENS_PROFILE": "standard" }
-    }
-  }
-}
+{
+  "mcpServers": {
+    "milens": {
+      "command": "milens",
+      "args": ["serve", "-p", "."],
+      "env": { "MILENS_PROFILE": "standard" }
+    }
+  }
+}

package/adapters/claude-code/.claude/settings.json.hooks-snippet.json

@@ -0,0 +1,26 @@
+{
+  "hooks": {
+    "SessionStart": [
+      {
+        "matcher": "*",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "milens hooks session-start --repo ${CLAUDE_PROJECT_DIR} || true"
+          }
+        ]
+      }
+    ],
+    "PreCompact": [
+      {
+        "matcher": "*",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "milens hooks pre-compact --repo ${CLAUDE_PROJECT_DIR} || true"
+          }
+        ]
+      }
+    ]
+  }
+}

package/adapters/claude-code/.claude-plugin/plugin.json

@@ -0,0 +1,11 @@
+{
+  "name": "milens",
+  "description": "AI-DOS Code Intelligence Platform — knowledge graph, security scanning, impact analysis, and closed-loop AI development with 43 MCP tools",
+  "version": "0.7.1",
+  "author": {
+    "name": "milens"
+  },
+  "homepage": "https://github.com/fuze210699/milens",
+  "repository": "https://github.com/fuze210699/milens",
+  "license": "MIT"
+}

package/adapters/claude-code/.mcp.json

@@ -0,0 +1,9 @@
+{
+  "mcpServers": {
+    "milens": {
+      "command": "milens",
+      "args": ["serve"],
+      "env": { "MILENS_PROFILE": "standard" }
+    }
+  }
+}

package/adapters/claude-code/CLAUDE.md

@@ -1,79 +1,79 @@
-# Milens — Code Intelligence for Claude Code
-
-## ⚠️ BEFORE ANYTHING ELSE
-
-This project is indexed by **milens**. You have MCP tools that are faster and more accurate than reading files directly.
-
-**HARD RULE: Use milens tools BEFORE built-in search/read tools.**
-
-| Situation | Built-in approach (❌) | Milens approach (✅) |
-|---|---|---|
-| Understand a function | Read 3-10 files | `mcp_milens_overview({name: "X"})` — 1 call |
-| Find references | grep in files | `mcp_milens_grep({pattern: "X"})` — searches everything |
-| Check edit safety | Hope nothing breaks | `mcp_milens_impact({target: "X"})` — exact blast radius |
-| Start working | Read README, explore files | `mcp_milens_codebase_summary()` — 500 token overview |
-| Before commit | `git diff` manually | `mcp_milens_detect_changes()` — symbols + risk scores |
-
-**If you skip milens tools, you're wasting tokens and risking breaking changes.**
-
-## ⭐ Core Tools (Use Every Session)
-
-| Tool | Purpose |
-|---|---|
-| `mcp_milens_overview` | **Use this first.** Context + impact + grep combined. 1 call replaces 3-5 file reads. |
-| `mcp_milens_impact` | Blast radius BEFORE editing. Shows what WILL BREAK. |
-| `mcp_milens_edit_check` | Pre-edit safety: callers, export status, re-export chains, test coverage |
-| `mcp_milens_context` | 360° view: incoming refs + outgoing deps |
-| `mcp_milens_query` | Find symbol definitions by name (camelCase/PascalCase/snake_case) |
-| `mcp_milens_grep` | Text search ALL project files (code, templates, docs, configs, styles) |
-| `mcp_milens_detect_changes` | Pre-commit: changed symbols + dependents + risk scores |
-| `mcp_milens_codebase_summary` | 500-token project overview. Use instead of reading README. |
-
-### 🔧 Situational Tools (Use When Needed)
-
-| Tool | Purpose | Use when... |
-|---|---|---|
-| `mcp_milens_guard_edit_check` | Hard pre-edit gate with audit tracking | Before every edit |
-| `mcp_milens_trace` | Execution flow from entrypoints | Debugging call chains |
-| `mcp_milens_explain_relationship` | Shortest path between two symbols | Understanding connections |
-| `mcp_milens_get_file_symbols` | All symbols in a file with ref/dep counts | Exploring a file |
-| `mcp_milens_get_type_hierarchy` | Inheritance/implementation tree | Class exploration |
-| `mcp_milens_find_dead_code` | Unused exported symbols | Before major refactors |
-| `mcp_milens_status` | Index health: symbols, links, files, coverage | Session start |
-
-All tool calls must include `repo` set to the absolute workspace root.
-
-## Session Workflow
-
-1. **Start** — The milens MCP server connects via `milens serve -p .`. Verify with `mcp_milens_status`.
-2. **Recall** — When asked to work on a symbol, use `mcp_milens_overview` for a combined view of context, impact, and text references.
-3. **Code** — Make changes following the edit-safety rules below.
-4. **Verify** — Before committing, run `mcp_milens_detect_changes` to confirm only expected files changed.
-5. **Annotate** — After significant changes, run `npx milens analyze -p . --force` to keep the index fresh.
-
-## Edit Safety (Mandatory)
-
-**Before editing any function, class, or method:**
-1. Call `mcp_milens_edit_check` on the target symbol to see callers and export status.
-2. If depth-1 dependents > 5, **stop and warn the user** before proceeding.
-3. Call `mcp_milens_context` to understand the full caller/callee picture.
-4. Only then make the edit.
-
-**Before deleting or renaming:**
-1. `mcp_milens_grep` — find ALL text references (templates, configs, docs).
-2. `mcp_milens_impact` with `direction: "upstream"` — find code-level dependents.
-3. Combine both results before acting.
-
-**Before committing:**
-- Call `mcp_milens_detect_changes` — if unexpected files appear, stop and report.
-
-## Tool Selection
-
-- Use `mcp_milens_query` for symbol names (camelCase, PascalCase, snake_case).
-- Use `mcp_milens_grep` for multi-word phrases, UI labels, error messages, or text in templates/configs/docs.
-- When in doubt, use `mcp_milens_grep` first — it searches everything.
-
-## Reference
-
-See `AGENTS.md` at the repo root for project-specific context and codebase structure.
-Skill files at `.agents/skills/milens/SKILL.md` provide advanced workflow guides.
+# Milens — Code Intelligence for Claude Code
+
+## ⚠️ BEFORE ANYTHING ELSE
+
+This project is indexed by **milens**. You have MCP tools that are faster and more accurate than reading files directly.
+
+**HARD RULE: Use milens tools BEFORE built-in search/read tools.**
+
+| Situation | Built-in approach (❌) | Milens approach (✅) |
+|---|---|---|
+| Understand a function | Read 3-10 files | `mcp_milens_overview({name: "X"})` — 1 call |
+| Find references | grep in files | `mcp_milens_grep({pattern: "X"})` — searches everything |
+| Check edit safety | Hope nothing breaks | `mcp_milens_impact({target: "X"})` — exact blast radius |
+| Start working | Read README, explore files | `mcp_milens_codebase_summary()` — 500 token overview |
+| Before commit | `git diff` manually | `mcp_milens_detect_changes()` — symbols + risk scores |
+
+**If you skip milens tools, you're wasting tokens and risking breaking changes.**
+
+## ⭐ Core Tools (Use Every Session)
+
+| Tool | Purpose |
+|---|---|
+| `mcp_milens_overview` | **Use this first.** Context + impact + grep combined. 1 call replaces 3-5 file reads. |
+| `mcp_milens_impact` | Blast radius BEFORE editing. Shows what WILL BREAK. |
+| `mcp_milens_edit_check` | Pre-edit safety: callers, export status, re-export chains, test coverage |
+| `mcp_milens_context` | 360° view: incoming refs + outgoing deps |
+| `mcp_milens_query` | Find symbol definitions by name (camelCase/PascalCase/snake_case) |
+| `mcp_milens_grep` | Text search ALL project files (code, templates, docs, configs, styles) |
+| `mcp_milens_detect_changes` | Pre-commit: changed symbols + dependents + risk scores |
+| `mcp_milens_codebase_summary` | 500-token project overview. Use instead of reading README. |
+
+### 🔧 Situational Tools (Use When Needed)
+
+| Tool | Purpose | Use when... |
+|---|---|---|
+| `mcp_milens_guard_edit_check` | Hard pre-edit gate with audit tracking | Before every edit |
+| `mcp_milens_trace` | Execution flow from entrypoints | Debugging call chains |
+| `mcp_milens_explain_relationship` | Shortest path between two symbols | Understanding connections |
+| `mcp_milens_get_file_symbols` | All symbols in a file with ref/dep counts | Exploring a file |
+| `mcp_milens_get_type_hierarchy` | Inheritance/implementation tree | Class exploration |
+| `mcp_milens_find_dead_code` | Unused exported symbols | Before major refactors |
+| `mcp_milens_status` | Index health: symbols, links, files, coverage | Session start |
+
+All tool calls must include `repo` set to the absolute workspace root.
+
+## Session Workflow
+
+1. **Start** — The milens MCP server connects via `milens serve -p .`. Verify with `mcp_milens_status`.
+2. **Recall** — When asked to work on a symbol, use `mcp_milens_overview` for a combined view of context, impact, and text references.
+3. **Code** — Make changes following the edit-safety rules below.
+4. **Verify** — Before committing, run `mcp_milens_detect_changes` to confirm only expected files changed.
+5. **Annotate** — After significant changes, run `npx milens analyze -p . --force` to keep the index fresh.
+
+## Edit Safety (Mandatory)
+
+**Before editing any function, class, or method:**
+1. Call `mcp_milens_edit_check` on the target symbol to see callers and export status.
+2. If depth-1 dependents > 5, **stop and warn the user** before proceeding.
+3. Call `mcp_milens_context` to understand the full caller/callee picture.
+4. Only then make the edit.
+
+**Before deleting or renaming:**
+1. `mcp_milens_grep` — find ALL text references (templates, configs, docs).
+2. `mcp_milens_impact` with `direction: "upstream"` — find code-level dependents.
+3. Combine both results before acting.
+
+**Before committing:**
+- Call `mcp_milens_detect_changes` — if unexpected files appear, stop and report.
+
+## Tool Selection
+
+- Use `mcp_milens_query` for symbol names (camelCase, PascalCase, snake_case).
+- Use `mcp_milens_grep` for multi-word phrases, UI labels, error messages, or text in templates/configs/docs.
+- When in doubt, use `mcp_milens_grep` first — it searches everything.
+
+## Reference
+
+See `AGENTS.md` at the repo root for project-specific context and codebase structure.
+Skill files at `.agents/skills/milens/SKILL.md` provide advanced workflow guides.

package/adapters/codex/.codex/config.toml

@@ -0,0 +1,3 @@
+[mcp_servers.milens]
+command = "milens"
+args = ["serve", "-p", "."]

package/adapters/copilot/.vscode/mcp.json

@@ -0,0 +1,10 @@
+{
+  "servers": {
+    "milens": {
+      "type": "stdio",
+      "command": "milens",
+      "args": ["serve", "-p", "${workspaceFolder}"],
+      "env": { "MILENS_PROFILE": "standard" }
+    }
+  }
+}

package/adapters/cursor/.cursor/mcp.json

@@ -0,0 +1,9 @@
+{
+  "mcpServers": {
+    "milens": {
+      "command": "milens",
+      "args": ["serve", "-p", "${workspaceFolder}"],
+      "env": { "MILENS_PROFILE": "standard" }
+    }
+  }
+}

package/adapters/cursor/.cursorrules

@@ -1,9 +1,69 @@
-{
-  "mcpServers": {
-    "milens": {
-      "command": "npx",
-      "args": ["milens", "serve"],
-      "env": { "MILENS_PROFILE": "standard" }
-    }
-  }
-}
+# Milens — Code Intelligence for Cursor
+
+## ⚠️ BEFORE ANYTHING ELSE
+
+This project is indexed by **milens**. You have MCP tools that are faster and more accurate than reading files directly.
+
+**HARD RULE: Use milens tools BEFORE built-in search/read tools.**
+
+| Situation | Built-in approach (❌) | Milens approach (✅) |
+|---|---|---|
+| Understand a function | Read 3-10 files | `mcp_milens_overview({name: "X"})` — 1 call |
+| Find references | grep in files | `mcp_milens_grep({pattern: "X"})` — searches everything |
+| Check edit safety | Hope nothing breaks | `mcp_milens_impact({target: "X"})` — exact blast radius |
+| Start working | Read README, explore files | `mcp_milens_codebase_summary()` — 500 token overview |
+| Before commit | `git diff` manually | `mcp_milens_detect_changes()` — symbols + risk scores |
+
+**If you skip milens tools, you're wasting tokens and risking breaking changes.**
+
+## MCP Connection
+
+Milens registers as an MCP server via `.cursor/mcp.json` (see that file in this adapter pack). All milens MCP calls require the `repo` parameter set to the absolute workspace root.
+
+## ⭐ Core Tools (Use Every Session)
+
+| Tool | Description |
+|---|---|
+| `mcp_milens_overview` | **Use this first.** Context + impact + grep combined. 1 call replaces 3-5 file reads. |
+| `mcp_milens_impact` | Blast radius BEFORE editing. Shows what WILL BREAK. |
+| `mcp_milens_edit_check` | Pre-edit safety: callers, export status, re-export chains, test coverage |
+| `mcp_milens_context` | Full incoming + outgoing dependency view |
+| `mcp_milens_query` | Symbol search by name |
+| `mcp_milens_grep` | Text search everywhere (templates, configs, docs, styles) |
+| `mcp_milens_detect_changes` | Pre-commit: changed symbols + dependents + risk scores |
+| `mcp_milens_codebase_summary` | 500-token project overview. Use instead of reading README. |
+
+### 🔧 Situational Tools (Use When Needed)
+
+| Tool | Description | Use when... |
+|---|---|---|
+| `mcp_milens_guard_edit_check` | Hard pre-edit gate with audit tracking | Before every edit |
+| `mcp_milens_trace` | Execution flow from entrypoints | Debugging call chains |
+| `mcp_milens_explain_relationship` | Shortest dependency path | Understanding connections |
+| `mcp_milens_get_type_hierarchy` | Class inheritance tree | Class exploration |
+| `mcp_milens_find_dead_code` | Unused exported symbols | Before major refactors |
+| `mcp_milens_status` | Index health: symbols, links, files, coverage | Session start |
+
+## Workflow
+
+**Every session follows this pattern:**
+
+1. **Start** — `mcp_milens_status` to verify the index is healthy.
+2. **Understand** — `mcp_milens_overview({name, repo})` for any symbol mentioned.
+3. **Edit safely** — `mcp_milens_edit_check` before touching hub/shared code. Halt if dependents > 5.
+4. **Verify** — `mcp_milens_detect_changes` before committing. Report unexpected changes.
+5. **Refresh** — `npx milens analyze -p . --force` after significant edits.
+
+## Mandatory Safety Rules
+
+- **Never edit** a symbol without `mcp_milens_edit_check` first.
+- **Never delete/rename** without `mcp_milens_grep` AND `mcp_milens_impact`.
+- **Never commit** without `mcp_milens_detect_changes`.
+- **Never omit** the `repo` parameter from milens calls.
+- **Use `mcp_milens_grep`** (not `mcp_milens_query`) for phrases, UI text, config strings.
+
+## Codebase Structure
+
+- `.agents/skills/milens/SKILL.md` — full milens documentation
+- `.agents/skills/` — area-specific skill files (analyzer, parser, server, store, etc.)
+- `AGENTS.md` — project-level instructions and conventions