sp-rag 0.6.11 → 0.6.13

package/README.md

@@ -17,7 +17,7 @@ CLI để setup nhanh SP-RAG theo hướng dev-friendly:
 ## Trạng thái package
 
 - package npm public: `sp-rag`
-- version đang publish: `0.6.7`
+- version đang publish: `0.6.13`
 - binary public: `sp-rag`
 
 ## Cài từ source trong monorepo
@@ -36,8 +36,10 @@ node dist/index.js doctor
 npx sp-rag@latest install --client codex --mcp-token <grc_pat_...> --doctor
 npx sp-rag@latest install --client cursor --scope project --cwd D:/Webs/seo-booster --mcp-token <grc_pat_...>
 npx sp-rag@latest install --client vscode --scope project --cwd D:/Webs/seo-booster --mcp-token <grc_pat_...>
-npx sp-rag@latest add --client claude-code --scope project --cwd D:/Webs/seo-booster
-npx sp-rag@latest explain --client vscode --scope project --cwd D:/Webs/seo-booster
+npx sp-rag@latest add --client claude-code --scope project --cwd D:/Webs/seo-booster
+npx sp-rag@latest update --client claude-code --scope project --cwd D:/Webs/seo-booster
+npx sp-rag@latest uninstall --client claude-code --scope project --cwd D:/Webs/seo-booster
+npx sp-rag@latest explain --client vscode --scope project --cwd D:/Webs/seo-booster
 npx sp-rag@latest token add --token <grc_pat_...>
 npx sp-rag@latest token verify --token <grc_pat_...>
 npx sp-rag@latest mcp add antigravity
@@ -50,8 +52,10 @@ Tương đương bằng `npm`:
 
 ```bash
 npm exec --yes sp-rag@latest install -- --client codex --mcp-token <grc_pat_...> --doctor
-npm exec --yes sp-rag@latest add -- --client claude-code --scope project --cwd D:/Webs/seo-booster
-npm exec --yes sp-rag@latest explain -- --client vscode --scope project --cwd D:/Webs/seo-booster
+npm exec --yes sp-rag@latest add -- --client claude-code --scope project --cwd D:/Webs/seo-booster
+npm exec --yes sp-rag@latest update -- --client claude-code --scope project --cwd D:/Webs/seo-booster
+npm exec --yes sp-rag@latest uninstall -- --client claude-code --scope project --cwd D:/Webs/seo-booster
+npm exec --yes sp-rag@latest explain -- --client vscode --scope project --cwd D:/Webs/seo-booster
 npm exec --yes sp-rag@latest token add -- --token <grc_pat_...>
 npm exec --yes sp-rag@latest token verify -- --token <grc_pat_...>
 ```
@@ -61,24 +65,29 @@ npm exec --yes sp-rag@latest token verify -- --token <grc_pat_...>
 Flow khuyên dùng sau khi đã có `grc_pat_*`:
 
 1. chạy `install` đúng một lần cho client đầu tiên, có kèm `--mcp-token`
-2. từ lần sau, dùng `add --client ...` để cài thêm MCP + skill cho client khác mà không phải nhập lại token
-3. khi chỉ muốn làm một nửa, dùng `mcp add` hoặc `skill install`
-4. khi đổi token, chỉ cần `token add`
-5. khi muốn kiểm tra máy đang được cấu hình ra sao, dùng `explain`
+2. từ lần sau, dùng `add --client ...` để cài thêm MCP + skill cho client khác mà không phải nhập lại token
+3. khi muốn cập nhật lại MCP + skill mà giữ token cũ, dùng `update`
+4. khi muốn gỡ sạch MCP + skill + config CLI do `sp-rag` tạo, dùng `uninstall`
+5. khi chỉ muốn làm một nửa, dùng `mcp add` hoặc `skill install`
+6. khi đổi token, chỉ cần `token add`
+7. khi muốn kiểm tra máy đang được cấu hình ra sao, dùng `explain`
 
 Ghi chú:
 
-- với `cursor` và `vscode` ở `scope project`, nếu Sếpp đang đứng sẵn trong repo thì có thể bỏ `--cwd`
-- CLI sẽ tự dùng thư mục hiện tại cho cả MCP lẫn skill
-- rule/agent mới đã được tăng độ ưu tiên MCP-first, giảm khả năng model nhảy thẳng sang grep/read file local
-- skill mới cũng dặn model tổng hợp từ `matched_passages`, `top_entities`, `top_relations`, `citations`; không bê nguyên `answer_brief`
+- với `cursor` và `vscode` ở `scope project`, nếu Sếpp đang đứng sẵn trong repo thì có thể bỏ `--cwd`
+- CLI sẽ tự dùng thư mục hiện tại cho cả MCP lẫn skill
+- `update` sẽ tự tính lại target project từ `--cwd` hoặc thư mục hiện tại, không dùng target cũ lệch project nếu user không truyền `--target-dir`
+- rule/agent mới route theo intent: feature/domain/docs thì MCP trước, còn edit/debug/current-code thì kiểm tra workspace trước
+- skill mới dùng hướng MCP-grounded + workspace-verified: tổng hợp từ `matched_passages`, `top_entities`, `top_relations`, `citations`, không bê nguyên `answer_brief`, và đối chiếu lại `git status --short`/file hiện tại khi cần
 
 Ví dụ:
 
 ```bash
-sp-rag install --client vscode --scope project --cwd D:/Webs/seo-booster --mcp-token <grc_pat_...> --doctor
-sp-rag add --client cursor --scope project --cwd D:/Webs/seo-booster
-sp-rag mcp add antigravity
+sp-rag install --client vscode --scope project --cwd D:/Webs/seo-booster --mcp-token <grc_pat_...> --doctor
+sp-rag add --client cursor --scope project --cwd D:/Webs/seo-booster
+sp-rag update --client vscode --scope project --cwd D:/Webs/seo-booster
+sp-rag uninstall --client claude-code --scope project --cwd D:/Webs/seo-booster
+sp-rag mcp add antigravity
 sp-rag skill install --client vscode --scope project --cwd D:/Webs/seo-booster
 sp-rag token add --token <grc_pat_moi>
 sp-rag explain --client vscode --scope project --cwd D:/Webs/seo-booster
@@ -186,7 +195,8 @@ Ghi chú:
 ### Skill / rule / custom agent
 
 - `codex`: `~/.codex/skills/sp-rag/SKILL.md`
-- `claude-code`: `~/.claude/skills/sp-rag/SKILL.md`
+- `claude-code` project: `.claude/skills/sp-rag/SKILL.md`
+- `claude-code` global skill-only override: `~/.claude/skills/sp-rag/SKILL.md`
 - `antigravity`: `~/.gemini/antigravity/skills/sp-rag/SKILL.md`
 - `opencode`: `~/.config/opencode/skills/sp-rag/SKILL.md`
 - `cursor` project: `.cursor/rules/sp-rag.mdc`
@@ -197,8 +207,10 @@ Ghi chú:
 
 ```bash
 sp-rag install --client codex --mcp-token <grc_pat_...> --doctor
-sp-rag add --client cursor --scope project --cwd D:/Webs/seo-booster
-sp-rag explain --client codex
+sp-rag add --client cursor --scope project --cwd D:/Webs/seo-booster
+sp-rag update --client cursor --scope project --cwd D:/Webs/seo-booster
+sp-rag uninstall --client cursor --scope project --cwd D:/Webs/seo-booster
+sp-rag explain --client codex
 sp-rag token add --token <grc_pat_...>
 sp-rag token verify --token <grc_pat_...>
 sp-rag config show
@@ -212,8 +224,10 @@ sp-rag mcp add vscode --scope project --cwd D:/Webs/seo-booster
 sp-rag mcp add opencode --scope project --cwd D:/Webs/seo-booster
 sp-rag skill install --client codex
 sp-rag skill install --client cursor --scope project --cwd D:/Webs/seo-booster
-sp-rag skill install --client vscode --scope project --cwd D:/Webs/seo-booster
-sp-rag eval run --file ./examples/eval-suite.sample.json
+sp-rag skill install --client vscode --scope project --cwd D:/Webs/seo-booster
+sp-rag eval run --file ./examples/eval-suite.sample.json
+sp-rag update --client vscode --scope project --cwd D:/Webs/seo-booster
+sp-rag uninstall --client vscode --scope project --cwd D:/Webs/seo-booster
 ```
 
 ## Lệnh chính
@@ -240,8 +254,10 @@ sp-rag mcp add opencode --scope project --cwd D:/Webs/seo-booster
 sp-rag skill install --client codex
 sp-rag skill install --client cursor --scope project --cwd D:/Webs/seo-booster
 sp-rag skill install --client vscode --scope project --cwd D:/Webs/seo-booster
-sp-rag eval run --file ./examples/eval-suite.sample.json
-sp-rag update setup --client codex
+sp-rag eval run --file ./examples/eval-suite.sample.json
+sp-rag update --client codex
+sp-rag update setup --client codex
+sp-rag uninstall --client codex
 ```
 
 ## `doctor` kiểm gì

package/dist/cli.js

@@ -1,10 +1,10 @@
 import { readFile } from 'node:fs/promises';
-import { loadCliConfig, saveCliConfig, } from './lib/config-store.js';
+import { clearCliConfig, loadCliConfig, saveCliConfig, } from './lib/config-store.js';
 import { runEvaluationSuite } from './lib/eval.js';
-import { defaultBaseUrl, defaultMcpServerAlias, defaultMcpUrl, installMcpConfig, resolveMcpConfigPath, } from './lib/mcp-config.js';
+import { defaultBaseUrl, defaultMcpServerAlias, defaultMcpUrl, installMcpConfig, removeMcpConfig, resolveMcpConfigPath, } from './lib/mcp-config.js';
 import { diagnoseMcpConfig, formatMcpDoctorResult } from './lib/doctor.js';
 import { fetchJson, fetchText, runDoctor } from './lib/http.js';
-import { installSkill, resolveSkillInstallTarget, } from './lib/skill.js';
+import { installSkill, removeSkill, resolveSkillInstallTarget, } from './lib/skill.js';
 let cliVersionPromise;
 async function resolveCliVersion() {
     if (!cliVersionPromise) {
@@ -205,9 +205,11 @@ function helpText() {
     return `sp-rag - CLI cho setup, MCP, codegraph, eval và skill của SP-RAG
 
 Lệnh chính:
-  sp-rag install --client codex|cursor|claude-code|antigravity|vscode|opencode [--base-url URL] [--mcp-url URL] [--scope global|project] [--mcp-token TOKEN] [--auth-env-var ENV_VAR] [--target-dir PATH] [--doctor]
-  sp-rag add --client codex|cursor|claude-code|antigravity|vscode|opencode [--scope global|project] [--cwd PATH] [--target-dir PATH]
-  sp-rag init [--base-url URL] [--mcp-url URL] [--client codex|cursor|claude-code|antigravity|vscode|opencode] [--skill-client codex|cursor|claude-code|antigravity|vscode|opencode] [--scope global|project] [--auth-env-var ENV_VAR] [--target-dir PATH]
+  sp-rag install --client codex|cursor|claude-code|antigravity|vscode|opencode [--base-url URL] [--mcp-url URL] [--scope global|project] [--mcp-token TOKEN] [--auth-env-var ENV_VAR] [--target-dir PATH] [--doctor]
+  sp-rag add --client codex|cursor|claude-code|antigravity|vscode|opencode [--scope global|project] [--cwd PATH] [--target-dir PATH]
+  sp-rag update [--client codex|cursor|claude-code|antigravity|vscode|opencode] [--scope global|project] [--cwd PATH] [--target-dir PATH] [--doctor]
+  sp-rag uninstall [--client codex|cursor|claude-code|antigravity|vscode|opencode] [--scope global|project] [--cwd PATH] [--target-dir PATH]
+  sp-rag init [--base-url URL] [--mcp-url URL] [--client codex|cursor|claude-code|antigravity|vscode|opencode] [--skill-client codex|cursor|claude-code|antigravity|vscode|opencode] [--scope global|project] [--auth-env-var ENV_VAR] [--target-dir PATH]
   sp-rag token add --token TOKEN
   sp-rag token verify [--token TOKEN] [--base-url URL]
   sp-rag config show
@@ -597,6 +599,50 @@ async function runUpdateSetup(parsed) {
         await runSkillInstall(parsed, defaults, client);
     }
 }
+async function runUpdate(parsed) {
+    const defaults = await loadRuntimeDefaults(parsed);
+    const client = resolveSelectedClient(parsed, defaults);
+    if (!client) {
+        throw new Error('Thiếu client. Dùng sp-rag update --client <client>.');
+    }
+    await runClientSetup(parsed, defaults, client);
+}
+async function runUninstall(parsed) {
+    const defaults = await loadRuntimeDefaults(parsed);
+    const client = resolveSelectedClient(parsed, defaults);
+    if (!client) {
+        throw new Error('Thiếu client. Dùng sp-rag uninstall --client <client>.');
+    }
+    const scope = supportedScope(optionString(parsed, 'scope')) ?? defaults.defaultScope;
+    const mcpResult = await removeMcpConfig({
+        client,
+        url: optionString(parsed, 'url') ?? defaults.mcpUrl,
+        scope,
+        cwd: optionString(parsed, 'cwd'),
+        serverAlias: optionString(parsed, 'server-alias') ?? defaults.serverAlias,
+    });
+    process.stdout.write(`${mcpResult.removed ? 'Đã gỡ' : 'Không tìm thấy'} cấu hình MCP cho ${mcpResult.client} tại ${mcpResult.path} (${mcpResult.scope}).\n`);
+    const skillClient = resolveSelectedSkillClient(parsed, defaults, client);
+    if (skillClient) {
+        const skillResult = await removeSkill({
+            client: skillClient,
+            cwd: optionString(parsed, 'cwd'),
+            scope,
+            targetDir: optionString(parsed, 'target-dir'),
+            serverAlias: optionString(parsed, 'server-alias') ?? defaults.serverAlias,
+            mcpUrl: optionString(parsed, 'mcp-url') ?? defaults.mcpUrl,
+            docsUrl: optionString(parsed, 'docs-url') ?? defaults.docsUrl,
+        });
+        for (const removedPath of skillResult.removedPaths) {
+            process.stdout.write(`Đã gỡ skill cho ${skillResult.client} tại ${removedPath}\n`);
+        }
+        if (skillResult.removedPaths.length === 0) {
+            process.stdout.write(`Không tìm thấy skill cần gỡ cho ${skillResult.client}.\n`);
+        }
+    }
+    const configPath = await clearCliConfig(defaults.homeDir);
+    process.stdout.write(`Đã xóa config CLI tại ${configPath}\n`);
+}
 export async function runCli(argv) {
     const parsed = parseArgv(argv);
     const [group, action] = parsed.positionals;
@@ -695,8 +741,17 @@ export async function runCli(argv) {
         if (group === 'eval' && action === 'run') {
             return runEval(parsed);
         }
-        if (group === 'update' && action === 'setup') {
-            await runUpdateSetup(parsed);
+        if (group === 'update' && (!action || action === 'setup')) {
+            if (action === 'setup') {
+                await runUpdateSetup(parsed);
+            }
+            else {
+                await runUpdate(parsed);
+            }
+            return 0;
+        }
+        if (group === 'uninstall') {
+            await runUninstall(parsed);
             return 0;
         }
         if (group === 'version') {

package/dist/lib/config-store.js

@@ -1,6 +1,6 @@
 import os from 'node:os';
 import path from 'node:path';
-import { mkdir, readFile, writeFile } from 'node:fs/promises';
+import { mkdir, readFile, rm, writeFile } from 'node:fs/promises';
 function stripUndefined(value) {
     return Object.fromEntries(Object.entries(value).filter(([, entry]) => entry !== undefined));
 }
@@ -37,3 +37,8 @@ export async function saveCliConfig(config, homeDir) {
     await writeFile(filePath, `${JSON.stringify(normalized, null, 2)}\n`, 'utf8');
     return filePath;
 }
+export async function clearCliConfig(homeDir) {
+    const filePath = resolveCliConfigPath(homeDir);
+    await rm(filePath, { force: true });
+    return filePath;
+}

package/dist/lib/mcp-config.js

@@ -85,6 +85,30 @@ function upsertObjectEntryPreservingAliasByEndpoint(base, sectionKey, entryKey,
         [sectionKey]: section,
     };
 }
+function removeObjectEntryByAliasOrEndpoint(base, sectionKey, entryKey, endpointKey, endpointValue) {
+    const section = base[sectionKey] && typeof base[sectionKey] === 'object'
+        ? { ...base[sectionKey] }
+        : {};
+    const targetEndpoint = normalizeEndpoint(endpointValue);
+    let removed = false;
+    for (const [alias, value] of Object.entries(section)) {
+        if (!value || typeof value !== 'object') {
+            continue;
+        }
+        const candidateEndpoint = normalizeEndpoint(value[endpointKey]);
+        if (alias === entryKey || (targetEndpoint && candidateEndpoint === targetEndpoint)) {
+            delete section[alias];
+            removed = true;
+        }
+    }
+    return {
+        next: {
+            ...base,
+            [sectionKey]: section,
+        },
+        removed,
+    };
+}
 function vscodeGlobalConfigPath(home) {
     if (process.platform === 'win32') {
         const appData = process.env['APPDATA']?.trim() || path.join(home, 'AppData', 'Roaming');
@@ -122,6 +146,23 @@ export function upsertCodexConfig(existing, options) {
     }
     return `${cleaned ? `${cleaned}\n\n` : ''}${lines.join('\n')}\n`;
 }
+export function removeCodexConfigContent(existing, serverAlias) {
+    const alias = serverAlias.trim();
+    const aliasPattern = escapeRegex(alias);
+    let removed = false;
+    const withoutHeaders = existing.replace(new RegExp(`(?:^|\\n)\\[mcp_servers\\.(?:"${aliasPattern}"|${aliasPattern})\\.headers\\][\\s\\S]*?(?=(?:\\n\\[[^\\n]+\\])|$)`, 'g'), () => {
+        removed = true;
+        return '';
+    });
+    const cleaned = withoutHeaders.replace(new RegExp(`(?:^|\\n)\\[mcp_servers\\.(?:"${aliasPattern}"|${aliasPattern})\\][\\s\\S]*?(?=(?:\\n\\[[^\\n]+\\])|$)`, 'g'), () => {
+        removed = true;
+        return '';
+    });
+    return {
+        content: `${cleaned.trimEnd()}${cleaned.trimEnd() ? '\n' : ''}`,
+        removed,
+    };
+}
 export function upsertJsonMcpConfig(existing, options) {
     const base = parseJsonObject(existing);
     const serverConfig = withHeaders({
@@ -279,3 +320,58 @@ export async function installMcpConfig(options) {
     await writeFile(resolved.path, content, 'utf8');
     return resolved;
 }
+export async function removeMcpConfig(options) {
+    const resolved = resolveMcpConfigPath(options);
+    const existing = await readFile(resolved.path, 'utf8').catch((error) => {
+        if (error.code === 'ENOENT') {
+            return '';
+        }
+        throw error;
+    });
+    const alias = options.serverAlias?.trim() || defaultMcpServerAlias();
+    const url = options.url.trim();
+    let content = existing;
+    let removed = false;
+    if (existing.trim()) {
+        switch (resolved.client) {
+            case 'codex': {
+                const result = removeCodexConfigContent(existing, alias);
+                content = result.content;
+                removed = result.removed;
+                break;
+            }
+            case 'cursor':
+            case 'claude-code': {
+                const result = removeObjectEntryByAliasOrEndpoint(parseJsonObject(existing), 'mcpServers', alias, 'url', url);
+                content = `${JSON.stringify(result.next, null, 2)}\n`;
+                removed = result.removed;
+                break;
+            }
+            case 'antigravity': {
+                const result = removeObjectEntryByAliasOrEndpoint(parseJsonObject(existing), 'mcpServers', alias, 'serverUrl', url);
+                content = `${JSON.stringify(result.next, null, 2)}\n`;
+                removed = result.removed;
+                break;
+            }
+            case 'vscode': {
+                const result = removeObjectEntryByAliasOrEndpoint(parseJsonObject(existing), 'servers', alias, 'url', url);
+                content = `${JSON.stringify(result.next, null, 2)}\n`;
+                removed = result.removed;
+                break;
+            }
+            case 'opencode': {
+                const result = removeObjectEntryByAliasOrEndpoint(parseJsonObject(existing), 'mcp', alias, 'url', url);
+                content = `${JSON.stringify(result.next, null, 2)}\n`;
+                removed = result.removed;
+                break;
+            }
+        }
+    }
+    if (removed) {
+        await writeFile(resolved.path, content, 'utf8');
+    }
+    return {
+        ...resolved,
+        removed,
+    };
+}

package/dist/lib/skill.js

@@ -1,6 +1,6 @@
 import os from 'node:os';
 import path from 'node:path';
-import { mkdir, writeFile } from 'node:fs/promises';
+import { mkdir, readFile, rm, writeFile } from 'node:fs/promises';
 const seoBoosterRemoteUrl = 'https://gitlab.com/secomapp/development/seo-booster.git';
 const seoBoosterProjectGuard = [
     '- SP-RAG chỉ dành cho repo seo-booster.',
@@ -72,7 +72,7 @@ export function defaultSkillDir(client = 'codex', cwd, scope) {
 function renderSkillMarkdown(context) {
     return `---
 name: sp-rag
-description: Use SP-RAG whenever the user asks about this codebase, internal business domain, rendered docs, import inventory, or codegraph sync status. You must call SP-RAG MCP tools first before answering from memory or reading local files.
+description: Use SP-RAG for seo-booster codebase, feature, domain, rendered-docs, import-inventory, or codegraph freshness work that needs grounded evidence.
 ---
 
 # SP-RAG
@@ -84,29 +84,43 @@ Docs URL: \`${context.docsUrl}\`
 
 ## When To Use This Skill
 
-- Use this skill whenever the request is about the internal codebase, business workflows, rendered docs, import inventory, or sync state.
-- Use this skill when grounded evidence matters more than memory or intuition.
-- Use this skill when the answer should come from MCP tools or rendered docs before freeform reasoning.
+- Use SP-RAG whenever the request is about seo-booster codebase, features, business workflows, rendered docs, import inventory, or sync state.
+- Use this skill for seo-booster codebase, feature, business workflow, rendered docs, import inventory, and sync-state questions.
+- The operating model is MCP-grounded, workspace-verified.
+- Do not treat MCP as the only source. MCP gives context; current workspace files and git state verify the final answer.
 
-## Recommended Workflow
+## Intent Routing
+
+- For feature, domain, architecture, workflow, docs, or "what does this do" questions: call \`${context.serverAlias}\` MCP first, usually \`query_context\`, then verify with workspace files when implementation details matter.
+- For current-code, edit, bug, failing-test, local-error, or "why is this broken right now" work: inspect the workspace first with \`git status --short\`, \`rg\`, and direct file reads, then use MCP to widen context if needed.
+- For mixed questions: get MCP context, inspect current files, compare both, then answer.
+- For docs questions, use \`get_rendered_docs\` when public, function, or dev docs may already answer the question.
+- For sync freshness and operations, use \`get_sync_status\`, \`get_sync_runs\`, and \`get_sync_metrics\`. Only use \`trigger_code_graph_sync\` when the user explicitly asks or stale evidence is the confirmed blocker.
+
+## Workspace Verification
+
+1. Run or mentally account for \`git status --short\` before relying on MCP for current behavior.
+2. If MCP cites files, inspect the current workspace version of those files when the answer depends on exact code.
+3. If changed files are related to the question, current workspace files override MCP evidence.
+4. If MCP and workspace disagree, say the graph may be stale and explain which source you are trusting.
+
+## Expand Evidence
 
-1. Call \`healthz\` if the MCP server might be unavailable or the connection is brand new.
-2. Use \`query_context\` for architecture, domain, entities, relations, and business flow questions.
-3. Use \`get_rendered_docs\` for public, function, or dev docs that were already rendered from the latest graph.
-4. Use \`get_sync_status\`, \`get_sync_runs\`, or \`get_sync_metrics\` when you need to verify commit freshness, investigate failures, or inspect operational history.
-5. Only call \`trigger_code_graph_sync\` when the user explicitly asks to refresh the graph or when a stale graph is the confirmed blocker.
-6. After \`query_context\` returns, synthesize the final answer from \`matched_passages\`, \`top_entities\`, \`top_relations\`, and \`citations\`. Treat \`answer_brief\` as a hint only.
+- After \`query_context\`, synthesize from \`matched_passages\`, \`top_entities\`, \`top_relations\`, \`citations\`, and \`feature_memory\` diagnostics when present.
+- Treat \`answer_brief\` as a hint only.
+- Do not stop at the first one or two MCP citations for broad feature questions.
+- Expand through related symbols, callers, services, controllers, jobs, routes, models, components, and tests using workspace search when needed.
+- Prefer rendered docs when they answer business behavior, but verify current code before making implementation claims.
 
 ## Guardrails
 
 ${seoBoosterProjectGuard}
-- You must call SP-RAG MCP tools first for codebase or domain questions before using local workspace search, grep, or file reads.
-- Prefer MCP-grounded answers before relying on memory.
-- Treat \`answer_brief\` as a hint only. Prefer the richer evidence in \`matched_passages\`, \`top_entities\`, \`top_relations\`, and \`citations\` when writing the final answer.
-- If the evidence may be stale, say so clearly and mention that the graph or docs may need a refresh.
+- Find root cause before fixing bugs.
+- Use evidence before conclusions; do not claim completion without verification.
+- For behavior changes, write or update tests before implementation when feasible.
+- If evidence may be stale, say so clearly and mention that the graph or docs may need a refresh.
 - Do not trigger sync or import actions unless the user asked for it or the workflow truly requires it.
-- When rendered docs already answer the question, cite or summarize those docs instead of rewriting everything from scratch.
-- Only fall back to local workspace search or file reads after MCP is unavailable or clearly lacks the needed evidence, and say that you are falling back.
+- When rendered docs already answer the question, cite or summarize those docs, then check current code if the answer depends on implementation details.
 `;
 }
 function renderCursorRule(context) {
@@ -119,14 +133,16 @@ alwaysApply: true
 # SP-RAG
 
 ${seoBoosterProjectGuard}
-- You must call the \`${context.serverAlias}\` MCP server first before using local workspace search, grep, or file reads for codebase and domain questions.
+- Route by intent before choosing tools.
+- Use the \`${context.serverAlias}\` MCP server first for feature, domain, architecture, workflow, docs, and "what does this do" questions.
+- Use workspace search first for current-code, edit, bug, and test-failure work.
+- Verify MCP evidence against \`git status --short\` and current workspace files before making implementation claims.
 - Use rendered docs from \`${context.docsUrl}\` when documentation already answers the question.
-- For architecture, domain, entities, relations, and business workflow questions, query MCP first and only then synthesize the answer.
 - Treat \`answer_brief\` only as a hint. Build the final answer from \`matched_passages\`, \`top_entities\`, \`top_relations\`, and \`citations\`.
+- Do not stop at the first one or two MCP citations for broad feature questions. Expand through related files, symbols, services, controllers, jobs, components, and tests.
 - For freshness or operational questions, check sync status, recent runs, and metrics before assuming the graph is current.
 - Only trigger codegraph sync when the user explicitly asks for it or stale evidence is the confirmed blocker.
 - If the evidence may be stale, say so clearly.
-- Only fall back to local workspace search after MCP is unavailable or clearly lacks the needed evidence, and say that you are falling back.
 `;
 }
 function renderVsCodeAgent(context) {
@@ -140,41 +156,44 @@ You are the SP-RAG custom agent for this workspace.
 
 Use the \`${context.serverAlias}\` MCP server at \`${context.mcpUrl}\` and the rendered docs URL \`${context.docsUrl}\` as your grounded source of truth.
 
-You must call an SP-RAG MCP tool before using local workspace search, grep, or file reads for codebase and domain questions.
+Route by intent before choosing tools. MCP gives the semantic map; the current workspace verifies exact code.
 
 ## Recommended Workflow
 
-1. Start with \`healthz\` if connectivity or freshness is uncertain.
-2. Use \`query_context\` for architecture, domain, entities, relations, and workflow questions.
+1. For feature, domain, architecture, workflow, docs, and "what does this do" questions, use \`query_context\` first.
+2. For current-code, edit, debug, failing-test, and local-error work, inspect the workspace first with \`git status --short\`, \`rg\`, and direct file reads.
 3. Use \`get_rendered_docs\` when public, function, or dev docs may already answer the question.
 4. Use \`get_sync_status\`, \`get_sync_runs\`, or \`get_sync_metrics\` for freshness, incident review, or operational debugging.
-5. Trigger codegraph sync only when the user explicitly requests a refresh or stale evidence is the confirmed blocker.
-6. After \`query_context\` returns, synthesize the final answer from \`matched_passages\`, \`top_entities\`, \`top_relations\`, and \`citations\`. Treat \`answer_brief\` as a hint.
+5. After MCP returns, verify important citations against current workspace files if the answer depends on exact implementation.
+6. Expand beyond the first one or two MCP citations for broad feature questions. Search related symbols, callers, services, controllers, jobs, routes, models, components, and tests.
+7. Trigger codegraph sync only when the user explicitly requests a refresh or stale evidence is the confirmed blocker.
 
 ## Guardrails
 
 ${seoBoosterProjectGuard}
-- Prefer MCP-grounded evidence before answering from memory.
+- Prefer MCP-grounded evidence before answering from memory, but current workspace files override MCP evidence when they differ.
 - Treat \`answer_brief\` as a hint. Prefer the richer evidence in \`matched_passages\`, \`top_entities\`, \`top_relations\`, and \`citations\` when writing the final answer.
-- Do not use local workspace search, grep, or file reads until SP-RAG MCP has been tried first for the current question.
 - When evidence may be stale, say so clearly and mention that a refresh might be needed.
 - Do not trigger sync or import actions unless the workflow truly requires it.
 - When rendered docs already answer the question, summarize those docs instead of rewriting everything from scratch.
-- Only fall back to local workspace search or file reads after MCP is unavailable or clearly lacks the needed evidence, and say that you are falling back.
+- Find root cause before fixes, write or update tests for behavior changes when feasible, and verify before claiming completion.
 `;
 }
 function renderVsCodeAlwaysOnInstructions(context) {
     return `# SP-RAG workspace instructions
 
 ${seoBoosterProjectGuard}
-- Always use the \`${context.serverAlias}\` MCP server first for codebase, domain, architecture, workflow, and operational questions before local workspace search, grep, or file reads.
+- Route by intent before choosing tools.
+- Use the \`${context.serverAlias}\` MCP server first for feature, domain, architecture, workflow, docs, and "what does this do" questions.
+- Use workspace inspection first for current-code, edit, debug, failing-test, and local-error work.
+- Verify MCP evidence against current workspace state, including \`git status --short\` and current file contents when implementation details matter.
 - Start with \`healthz\` only when connectivity is uncertain or the session is new.
-- Use \`query_context\` first for feature, entity, relation, and business-flow questions.
 - When \`query_context\` returns, synthesize the final answer from \`matched_passages\`, \`top_entities\`, \`top_relations\`, and \`citations\`.
 - Treat \`answer_brief\` as a hint only, not the final answer.
+- Expand beyond the first one or two MCP citations for broad feature questions.
 - Use \`get_rendered_docs\` when rendered docs can answer the question faster than raw code inspection.
 - Only trigger codegraph sync when the user explicitly asks for it or stale graph evidence is the confirmed blocker.
-- If MCP is unavailable or clearly lacks evidence, say so explicitly before falling back to local workspace search or file reads.
+- If MCP and current workspace files disagree, say MCP may be stale and trust the current workspace for implementation details.
 `;
 }
 function renderSkillArtifact(context) {
@@ -260,6 +279,41 @@ export async function installSkill(options = {}) {
         paths,
     };
 }
+async function removeIfSpRagArtifact(filePath) {
+    const content = await readFile(filePath, 'utf8').catch((error) => {
+        if (error.code === 'ENOENT') {
+            return null;
+        }
+        throw error;
+    });
+    if (content === null) {
+        return false;
+    }
+    if (!content.includes('SP-RAG') && !content.includes('sp-rag')) {
+        return false;
+    }
+    await rm(filePath, { force: true });
+    return true;
+}
+export async function removeSkill(options = {}) {
+    const resolved = resolveSkillInstallTarget(options);
+    const context = createSkillContext(options, resolved.client);
+    const paths = [
+        resolved.path,
+        ...resolveAdditionalSkillArtifacts(options, resolved, context).map((entry) => entry.path),
+    ];
+    const removedPaths = [];
+    for (const filePath of paths) {
+        if (await removeIfSpRagArtifact(filePath)) {
+            removedPaths.push(filePath);
+        }
+    }
+    return {
+        client: resolved.client,
+        paths,
+        removedPaths,
+    };
+}
 export async function installCodexSkill(options = {}) {
     return installSkill({
         ...options,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sp-rag",
-  "version": "0.6.11",
+  "version": "0.6.13",
   "description": "CLI cho setup MCP, codegraph GitNexus và skill của SP-RAG",
   "type": "module",
   "files": [