viepilot 2.50.1 → 3.7.2

package/skills/vp-docs/SKILL.md

@@ -53,7 +53,7 @@ Silent if command unavailable or errors.
 </persona_context>
 
 
-<cursor_skill_adapter>
+<adapter id="claude-code">
 ## A. Skill Invocation
 - Skill được gọi khi user mention `vp-docs`, `/vp-docs`, "docs", "documentation", "tài liệu"
 - Treat all user text after the skill mention as `{{VP_ARGS}}`
@@ -62,8 +62,57 @@ Silent if command unavailable or errors.
 Prompt user conversationally with options.
 
 ## C. Tool Usage
-Use Cursor tools: `Shell`, `ReadFile`, `Glob`, `rg`, `ApplyPatch`, `WebSearch`, `WebFetch`, `Subagent`
-</cursor_skill_adapter>
+Use Claude Code tools: `Bash` (shell), `Read` (file), `Edit` + `Write` (file write/patch),
+`Grep` (search), `Glob` (file patterns), `LS`, `WebSearch`, `WebFetch`,
+`Agent` (spawn subagent — multi-level nesting supported)
+Interactive: `AskUserQuestion` (deferred — preload via ToolSearch before first call)
+</adapter>
+
+<adapter id="cursor-agent">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+
+## C. Tool Usage
+Use Cursor tools: `run_terminal_cmd` (shell), `read_file` (read), `edit_file` (write/edit),
+`grep_search` (search), `web_search`, `codebase_search`, `list_dir`, `file_search`
+Interactive: text list fallback (AskQuestion available in Plan Mode only; Agent Mode = text)
+Subagent: `/multitask` (user command, single-level only — not a callable tool)
+MCP limit: 40 tools
+</adapter>
+
+<adapter id="antigravity">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+Skill discovery: LLM-driven (automatic, no slash command needed).
+
+## C. Tool Usage
+Use Antigravity tools: `shell` (cmd), `file_read`, `file_write`, MCP plugins
+Interactive: text fallback (TUI-based; no formal AskUserQuestion)
+Skill path: `.agents/skills/<skill>/SKILL.md` (project) or `~/.gemini/antigravity/skills/` (global)
+Note: Gemini CLI deprecated June 18, 2026 — use Antigravity CLI.
+</adapter>
+
+<adapter id="codex">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+
+## C. Tool Usage
+Use Codex tools: `container.exec` (sandboxed shell), `apply_patch` (file write), `web_search`
+Interactive: text fallback (TUI Tab/Enter injection)
+Config: `~/.codex/config.toml`
+</adapter>
+
+<adapter id="copilot">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+Discovery: User-driven (`@agent-name` in GitHub Copilot Chat).
+
+## C. Tool Usage
+Use Copilot tools: `runCommands` (shell), `read`/`readfile` (read), `edit`/`editFiles` (write),
+`code_search`, `find_references`
+Interactive: `askQuestions` (main agent only — NOT available in subagents; VS Code issue #293745)
+Skill path: `.github/agents/<name>.agent.md`
+</adapter>
 <scope_policy>
 ## ViePilot Namespace Guard (BUG-004)
 - Default mode: only use and reference `vp-*` skills in ViePilot workflows.

package/skills/vp-evolve/SKILL.md

@@ -53,7 +53,7 @@ Silent if command unavailable or errors.
 </persona_context>
 
 
-<cursor_skill_adapter>
+<adapter id="claude-code">
 ## A. Skill Invocation
 - Skill được gọi khi user mention `vp-evolve`, `/vp-evolve`, "evolve", "thêm feature", "milestone mới", "upgrade"
 - Treat all user text after the skill mention as `{{VP_ARGS}}`
@@ -62,8 +62,57 @@ Silent if command unavailable or errors.
 Prompt user conversationally with options.
 
 ## C. Tool Usage
-Use Cursor tools: `Shell`, `ReadFile`, `Glob`, `rg`, `ApplyPatch`, `WebSearch`, `WebFetch`, `Subagent`
-</cursor_skill_adapter>
+Use Claude Code tools: `Bash` (shell), `Read` (file), `Edit` + `Write` (file write/patch),
+`Grep` (search), `Glob` (file patterns), `LS`, `WebSearch`, `WebFetch`,
+`Agent` (spawn subagent — multi-level nesting supported)
+Interactive: `AskUserQuestion` (deferred — preload via ToolSearch before first call)
+</adapter>
+
+<adapter id="cursor-agent">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+
+## C. Tool Usage
+Use Cursor tools: `run_terminal_cmd` (shell), `read_file` (read), `edit_file` (write/edit),
+`grep_search` (search), `web_search`, `codebase_search`, `list_dir`, `file_search`
+Interactive: text list fallback (AskQuestion available in Plan Mode only; Agent Mode = text)
+Subagent: `/multitask` (user command, single-level only — not a callable tool)
+MCP limit: 40 tools
+</adapter>
+
+<adapter id="antigravity">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+Skill discovery: LLM-driven (automatic, no slash command needed).
+
+## C. Tool Usage
+Use Antigravity tools: `shell` (cmd), `file_read`, `file_write`, MCP plugins
+Interactive: text fallback (TUI-based; no formal AskUserQuestion)
+Skill path: `.agents/skills/<skill>/SKILL.md` (project) or `~/.gemini/antigravity/skills/` (global)
+Note: Gemini CLI deprecated June 18, 2026 — use Antigravity CLI.
+</adapter>
+
+<adapter id="codex">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+
+## C. Tool Usage
+Use Codex tools: `container.exec` (sandboxed shell), `apply_patch` (file write), `web_search`
+Interactive: text fallback (TUI Tab/Enter injection)
+Config: `~/.codex/config.toml`
+</adapter>
+
+<adapter id="copilot">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+Discovery: User-driven (`@agent-name` in GitHub Copilot Chat).
+
+## C. Tool Usage
+Use Copilot tools: `runCommands` (shell), `read`/`readfile` (read), `edit`/`editFiles` (write),
+`code_search`, `find_references`
+Interactive: `askQuestions` (main agent only — NOT available in subagents; VS Code issue #293745)
+Skill path: `.github/agents/<name>.agent.md`
+</adapter>
 <scope_policy>
 ## ViePilot Namespace Guard (BUG-004)
 - Default mode: only use and reference `vp-*` skills in ViePilot workflows.

package/skills/vp-info/SKILL.md

@@ -53,14 +53,63 @@ Silent if command unavailable or errors.
 </persona_context>
 
 
-<cursor_skill_adapter>
+<adapter id="claude-code">
 ## A. Skill Invocation
 - Skill được gọi khi user mention `vp-info`, `/vp-info`, "viepilot version", "phiên bản viepilot", "skills list bundle"
 - Treat all user text after the skill mention as `{{VP_ARGS}}`
 
 ## B. Tool Usage
-Use Cursor tools: `Shell`, `ReadFile`, `Glob`, `rg`, `ApplyPatch`, `WebSearch`, `WebFetch`, `Subagent`
-</cursor_skill_adapter>
+Use Claude Code tools: `Bash` (shell), `Read` (file), `Edit` + `Write` (file write/patch),
+`Grep` (search), `Glob` (file patterns), `LS`, `WebSearch`, `WebFetch`,
+`Agent` (spawn subagent — multi-level nesting supported)
+Interactive: `AskUserQuestion` (deferred — preload via ToolSearch before first call)
+</adapter>
+
+<adapter id="cursor-agent">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+
+## C. Tool Usage
+Use Cursor tools: `run_terminal_cmd` (shell), `read_file` (read), `edit_file` (write/edit),
+`grep_search` (search), `web_search`, `codebase_search`, `list_dir`, `file_search`
+Interactive: text list fallback (AskQuestion available in Plan Mode only; Agent Mode = text)
+Subagent: `/multitask` (user command, single-level only — not a callable tool)
+MCP limit: 40 tools
+</adapter>
+
+<adapter id="antigravity">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+Skill discovery: LLM-driven (automatic, no slash command needed).
+
+## C. Tool Usage
+Use Antigravity tools: `shell` (cmd), `file_read`, `file_write`, MCP plugins
+Interactive: text fallback (TUI-based; no formal AskUserQuestion)
+Skill path: `.agents/skills/<skill>/SKILL.md` (project) or `~/.gemini/antigravity/skills/` (global)
+Note: Gemini CLI deprecated June 18, 2026 — use Antigravity CLI.
+</adapter>
+
+<adapter id="codex">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+
+## C. Tool Usage
+Use Codex tools: `container.exec` (sandboxed shell), `apply_patch` (file write), `web_search`
+Interactive: text fallback (TUI Tab/Enter injection)
+Config: `~/.codex/config.toml`
+</adapter>
+
+<adapter id="copilot">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+Discovery: User-driven (`@agent-name` in GitHub Copilot Chat).
+
+## C. Tool Usage
+Use Copilot tools: `runCommands` (shell), `read`/`readfile` (read), `edit`/`editFiles` (write),
+`code_search`, `find_references`
+Interactive: `askQuestions` (main agent only — NOT available in subagents; VS Code issue #293745)
+Skill path: `.github/agents/<name>.agent.md`
+</adapter>
 <scope_policy>
 ## ViePilot Namespace Guard (BUG-004)
 - Default mode: only use and reference `vp-*` skills in ViePilot workflows.

package/skills/vp-intake/SKILL.md

@@ -37,7 +37,7 @@ Inject the output as `## User Persona` context before any task execution.
 Silent if command unavailable or errors.
 </persona_context>
 
-<cursor_skill_adapter>
+<adapter id="claude-code">
 ## A. Skill Invocation
 - Skill được gọi khi user mention `vp-intake`, `/vp-intake`, "import tickets", "nhập ticket", "đọc ticket từ", "triage ticket"
 - Treat all user text after the skill mention as `{{VP_ARGS}}`
@@ -46,8 +46,57 @@ Silent if command unavailable or errors.
 Prompt user conversationally with options.
 
 ## C. Tool Usage
-Use Cursor tools: `Shell`, `ReadFile`, `Glob`, `rg`, `ApplyPatch`, `WebSearch`, `WebFetch`, `Subagent`
-</cursor_skill_adapter>
+Use Claude Code tools: `Bash` (shell), `Read` (file), `Edit` + `Write` (file write/patch),
+`Grep` (search), `Glob` (file patterns), `LS`, `WebSearch`, `WebFetch`,
+`Agent` (spawn subagent — multi-level nesting supported)
+Interactive: `AskUserQuestion` (deferred — preload via ToolSearch before first call)
+</adapter>
+
+<adapter id="cursor-agent">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+
+## C. Tool Usage
+Use Cursor tools: `run_terminal_cmd` (shell), `read_file` (read), `edit_file` (write/edit),
+`grep_search` (search), `web_search`, `codebase_search`, `list_dir`, `file_search`
+Interactive: text list fallback (AskQuestion available in Plan Mode only; Agent Mode = text)
+Subagent: `/multitask` (user command, single-level only — not a callable tool)
+MCP limit: 40 tools
+</adapter>
+
+<adapter id="antigravity">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+Skill discovery: LLM-driven (automatic, no slash command needed).
+
+## C. Tool Usage
+Use Antigravity tools: `shell` (cmd), `file_read`, `file_write`, MCP plugins
+Interactive: text fallback (TUI-based; no formal AskUserQuestion)
+Skill path: `.agents/skills/<skill>/SKILL.md` (project) or `~/.gemini/antigravity/skills/` (global)
+Note: Gemini CLI deprecated June 18, 2026 — use Antigravity CLI.
+</adapter>
+
+<adapter id="codex">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+
+## C. Tool Usage
+Use Codex tools: `container.exec` (sandboxed shell), `apply_patch` (file write), `web_search`
+Interactive: text fallback (TUI Tab/Enter injection)
+Config: `~/.codex/config.toml`
+</adapter>
+
+<adapter id="copilot">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+Discovery: User-driven (`@agent-name` in GitHub Copilot Chat).
+
+## C. Tool Usage
+Use Copilot tools: `runCommands` (shell), `read`/`readfile` (read), `edit`/`editFiles` (write),
+`code_search`, `find_references`
+Interactive: `askQuestions` (main agent only — NOT available in subagents; VS Code issue #293745)
+Skill path: `.github/agents/<name>.agent.md`
+</adapter>
 <scope_policy>
 ## ViePilot Namespace Guard (BUG-004)
 - Default mode: only use and reference `vp-*` skills in ViePilot workflows.
@@ -80,6 +129,11 @@ Optional flags:
 - `--dry-run` : Classify and show tickets without creating requests or writing back
 - `--setup` : Force setup wizard — configure a new channel now (even if channels already exist)
 - `--config` : Alias for `--setup`
+- `--skip-validation` : Skip Step 4.5 codebase validation (faster, no file-scanner-agent spawn)
+- `--analyze` : Force re-analysis of source file structure via `analyze_structure` op, even if a fresh manifest exists (ENH-095)
+- `--schedule "CRON"` : Create a CronCreate schedule for automated intake (Claude Code only)
+- `--unschedule` : Delete the existing intake schedule (CronDelete)
+- `--auto` : Run in headless auto mode — set by scheduler only, not for manual use
 
 **Supported channel types:**
 | Type | Auth | Config field |
@@ -88,6 +142,11 @@ Optional flags:
 | `google_sheets` | Service Account JSON | `spreadsheet_id` + `sheet_name` |
 | `excel_m365` (Graph API) | Azure App Registration | `workbook_id` + `sheet_name` |
 | `excel_m365` (sharing link) | None — anonymous | `sharing_url` |
+| `browser` | None — public URL only | `url` (https://...) |
+
+**`browser` channel** — reads any publicly accessible URL via `vercel-labs/agent-browser`.
+Supports: Google Sheets share links, GitHub Issues lists, Jira public boards, Trello boards, Notion pages.
+Prerequisite (Claude Code only): `npx skills add vercel-labs/agent-browser`
 
 **Config file:** `.viepilot/intake/channels.json`
 **Credentials dir:** `.viepilot/.credentials/` (gitignored)
@@ -95,6 +154,76 @@ Optional flags:
 
 <process>
 
+### --schedule flag (ENH-088, Claude Code only)
+
+```js
+// --schedule "0 9 * * 1"
+// ToolSearch("select:CronCreate") must be called first (deferred tool)
+CronCreate({ schedule: cronExpression, prompt: `/vp-intake --channel ${channelId} --auto` })
+// Then write schedule.json via lib/intake/auto-intake.cjs createSchedule()
+```
+Write `.viepilot/intake/schedule.json` with `{ cron, schedule_id, channel_id, created }`.
+Output: `✅ Intake scheduled: ${humanReadableCron(cronExpression)}`
+
+**Non-CC adapters**: print the following and exit:
+```
+Scheduling requires Claude Code. Use cron + CLI instead.
+Suggested: cron "0 9 * * 1 claude /vp-intake --channel default --auto"
+```
+
+### --unschedule flag (ENH-088)
+
+```js
+// ToolSearch("select:CronDelete") must be called first
+const { schedule_id } = readSchedule()   // lib/intake/auto-intake.cjs
+CronDelete({ schedule_id })
+deleteSchedule(projectRoot)              // removes schedule.json
+```
+Output: `✅ Intake schedule removed`
+If no schedule.json exists: `No intake schedule found.`
+
+### --auto mode (ENH-088, headless — set by scheduler only)
+
+When `--auto` flag is present:
+1. Skip Step 0 setup wizard, Step 3 channel select (use `--channel` arg), Step 5 AUQ triage
+2. Call `runAutoIntake(tickets, channel, { projectRoot })` from `lib/intake/auto-intake.cjs`
+3. Auto-accept tickets with `confidence ≥ 0.7` → create request files
+4. Queue lower-confidence tickets → append to `.viepilot/intake/pending-review.json`
+5. Print auto-run summary and exit
+
+Output format:
+```
+[AUTO INTAKE] Channel: {name} | {N} tickets classified
+Auto-accepted ({N}): BUG-030, ENH-090 (confidence ≥ 0.7)
+Queued for review ({N}): TICKET-007 (confidence 0.4)
+Report: .viepilot/intake/TRIAGE-auto-{timestamp}.md
+```
+
+### Pending Review Queue (ENH-088, manual runs only)
+
+At the start of every manual `/vp-intake` (before Step 0), check for queued items:
+
+```js
+const { readSchedule } = require('lib/intake/auto-intake.cjs');
+const pendingPath = '.viepilot/intake/pending-review.json';
+if (fs.existsSync(pendingPath)) {
+  const pending = JSON.parse(fs.readFileSync(pendingPath));
+  if (pending.items && pending.items.length > 0) {
+    // Show AUQ:
+    const lastDate = pending.items[pending.items.length - 1].queued_at?.split('T')[0];
+    AskUserQuestion({
+      question: `⚠️ ${pending.items.length} tickets in pending-review queue (from last auto-run ${lastDate}). Process them now?`,
+      options: [
+        { label: 'Yes', description: 'Load pending items into triage' },
+        { label: 'Skip', description: 'Proceed with fresh intake' },
+      ]
+    })
+    // If Yes: inject pending.items into the ticket list for Step 4 onwards
+    // If Skip: continue normally
+  }
+}
+```
+
 ### Step 0: Setup wizard detection (ENH-084)
 
 **Check wizard trigger conditions:**
@@ -150,15 +279,110 @@ header: "Channel"
 options: one per channel — label: "{channel.name} ({channel.type})", description: "{channel.id}"
 ```
 
+### Step 3.5: Sharing URL guard (ENH-089)
+
+Before reading, check for `sharing_url` on `excel_m365` channels:
+```
+if (channel.type === 'excel_m365' && channel.sharing_url) {
+  // Print warning — write-back will be disabled for this session
+  console.warn(
+    `⚠️  Channel "${channel.name}" uses a sharing_url — write-back is read-only for sharing links.\n` +
+    `   To enable write-back: configure workbook_id + .viepilot/.credentials/m365-credentials.json`
+  );
+}
+```
+Continue with intake — read-only is acceptable for triage.
+
+### Step 3.6: Manifest Lifecycle (ENH-095) — Claude Code only
+
+Before reading tickets, check for an existing intake manifest and use it for column mapping.
+Non-CC adapters skip this step and fall back to `autoDetectColumnMap()` as before.
+
+```js
+const { loadManifest, isManifestFresh, saveManifest, getColumnMap } = require('lib/intake/manifest.cjs');
+
+const channelId = channel.id || channel.name.toLowerCase().replace(/\s+/g, '-');
+const ttlDays   = config?.intake?.manifest_ttl_days ?? 7;
+const existing  = loadManifest(channelId, projectRoot);
+const forceAnalyze = flags.includes('--analyze');
+
+let manifest = null;
+
+if (existing && isManifestFresh(existing, ttlDays) && !forceAnalyze) {
+  // Use cached manifest — skip re-analysis
+  manifest = existing;
+  console.log(`[vp-intake] Manifest: .viepilot/intake/${channelId}-manifest.json (${existing.analyzed_at})`);
+
+} else if (channel.type === 'excel_m365' || channel.type === 'google_sheets') {
+  // Dispatch analyze_structure — AI reads entire file and maps structure
+  const agentType = channel.type === 'excel_m365' ? 'excel-intake-agent' : 'sheets-intake-agent';
+  const result = await Agent({
+    subagent_type: agentType,
+    description: `${agentType}: analyze_structure`,
+    prompt: `op: analyze_structure. channel_config: ${JSON.stringify(channel)}. projectRoot: ${projectRoot}`
+  });
+  if (result && !result.error) {
+    manifest = typeof result === 'string' ? JSON.parse(result) : result;
+    saveManifest(channelId, manifest, projectRoot);
+    console.log(`[vp-intake] Manifest saved → .viepilot/intake/${channelId}-manifest.json`);
+  }
+}
+
+// Override column_map from manifest when not explicitly configured
+if (manifest && !channel.column_map) {
+  const detectedMap = getColumnMap(manifest, channel.sheet_name || null);
+  if (detectedMap && Object.keys(detectedMap).length > 0) {
+    channel = { ...channel, column_map: detectedMap };
+    console.log(`[vp-intake] Column map from manifest: ${JSON.stringify(detectedMap)}`);
+  }
+}
+```
+
+**`--analyze` flag**: forces re-analysis even if manifest is fresh (e.g., source file structure changed).
+
 ### Step 4: Read and classify tickets
 
-Dispatch to the correct adapter based on `channel.type`:
+**Claude Code adapter** — dispatch via native agents for Excel/Sheets/Browser:
+```js
+if (channel.type === 'excel_m365') {
+  // Claude Code: use native excel-intake-agent
+  Agent({ subagent_type: "excel-intake-agent",
+    description: "excel-intake-agent: read tickets from Excel M365",
+    prompt: `op: read. channel_config: ${JSON.stringify(channel)}. projectRoot: ${projectRoot}`
+  })
+} else if (channel.type === 'google_sheets') {
+  // Claude Code: use native sheets-intake-agent
+  Agent({ subagent_type: "sheets-intake-agent",
+    description: "sheets-intake-agent: read tickets from Google Sheets",
+    prompt: `op: read. channel_config: ${JSON.stringify(channel)}. projectRoot: ${projectRoot}`
+  })
+} else if (channel.type === 'browser') {
+  // SharePoint xlsx sharing links → excel-intake-agent (ENH-094)
+  const { detectUrlType } = require('lib/intake/adapters/browser.cjs');
+  if (detectUrlType(channel.url) === 'sharepoint-xlsx') {
+    Agent({ subagent_type: "excel-intake-agent",
+      description: "excel-intake-agent: read SharePoint xlsx via sharing_url",
+      prompt: `op: read. channel_config: ${JSON.stringify({ ...channel, type: 'excel_m365', sharing_url: channel.url })}. projectRoot: ${projectRoot}`
+    })
+  } else {
+    // All other browser URLs → browser-intake-agent (requires vercel-labs/agent-browser)
+    Agent({ subagent_type: "browser-intake-agent",
+      description: "browser-intake-agent: read tickets from public URL",
+      prompt: `op: read_url. url: ${channel.url}. channel_config: ${JSON.stringify(channel)}. projectRoot: ${projectRoot}`
+    })
+  }
+}
+```
+
+**Non-CC adapters** — inline adapter dispatch (unchanged):
 - `csv` → `lib/intake/adapters/csv.cjs` → `readCsv(channel, projectRoot)`
 - `google_sheets` → `lib/intake/adapters/google-sheets.cjs` → `readGoogleSheet(channel, projectRoot)`
 - `excel_m365` → `lib/intake/adapters/excel-m365.cjs` → `readExcelM365(channel, projectRoot)`
+- `browser` → `lib/intake/adapters/browser.cjs` → `readBrowserUrl(channel, projectRoot)` (throws — CC only)
 
 For each ticket, call `classifyTicket(ticket)` from `lib/intake/classifier.cjs`.
-Attach `ticket._classified = 'BUG' | 'ENH' | 'UNCLEAR'`.
+Returns `{ classified: 'BUG'|'ENH'|'UNCLEAR', confidence: 0.0–1.0 }`.
+Attach `ticket._classified = result.classified` and `ticket._confidence = result.confidence`.
 
 Display classification summary:
 ```
@@ -168,6 +392,35 @@ Read {N} tickets from {channel.name}
 
 If 0 tickets found, exit with message "No tickets found in this channel."
 
+### Step 4.5: Parallel Codebase Validation (ENH-087)
+
+> Skip entirely if `--skip-validation` flag is set.
+
+Call `validateTickets(classifiedTickets, projectRoot)` from `lib/intake/validator.cjs`.
+Fan-out: one validation per ticket (batched ≤10 concurrently via `Promise.all`).
+
+**Claude Code adapter** — each ticket dispatches a `file-scanner-agent`:
+```js
+// Inside validateTicket (CC path override):
+Agent({ subagent_type: "file-scanner-agent",
+  description: `file-scanner-agent: validate "${ticket.title}"`,
+  prompt: `keywords: ${JSON.stringify(keywords)}. projectRoot: ${projectRoot}. requests_dir: .viepilot/requests/`
+})
+```
+
+**Non-CC adapters** — inline `grep` + request scan (no agent spawn).
+
+Display progress during validation:
+```
+Validating {N} tickets against codebase... (parallel)
+  ✅ "Login button broken"      — found: src/components/LoginButton.tsx
+  ⚠️ "Add CSV export"           — similar: ENH-045
+  ❓ "Fix the performance issue" — no codebase match
+```
+
+Validation results are attached as `ticket._validation` and shown as badges in Step 5 AUQ.
+`--skip-validation` flag: skip Step 4.5 entirely, proceed directly to Step 5.
+
 ### Step 5: Triage (AUQ multi-select)
 
 Call `runTriage(tickets, channel, projectRoot, askFn)` from `lib/intake/triage-ux.cjs`.
@@ -182,10 +435,54 @@ async function askFn(question, options, multiSelect) {
 For each ticket: AUQ multi-select to accept/decline, then AUQ single-select for decline reason.
 UNCLEAR tickets get a 3-choice prompt: "Accept as BUG / Accept as ENH / Decline".
 
+### Step 5.5: Embed Intake Source in accepted-ticket request files (ENH-095)
+
+When creating `.viepilot/requests/BUG-N.md` or `ENH-N.md` for an accepted ticket,
+append the following block so `vp-auto` can write back task completion to the source row:
+
+```markdown
+## Intake Source
+- channel_id:    {channelId}
+- sheet_name:    {channel.sheet_name || null}
+- source_row:    {ticket._source_row}
+- manifest_path: .viepilot/intake/{channelId}-manifest.json
+- channel_type:  {channel.type}
+- workbook_id:   {channel.workbook_id || null}
+- sharing_url:   {channel.sharing_url || null}
+```
+
+This block is read by `vp-auto` post-PASS hook to call `writebackIntakeResponse()`.
+For read-only channels (`sharing_url`, `browser`, `csv`), write-back will be skipped
+silently — the block is still written for traceability.
+
 ### Step 6: Write-back + Report
 
+**Claude Code adapter** — dispatch via native agents for Excel/Sheets write-back:
+```js
+if (channel.type === 'excel_m365') {
+  Agent({ subagent_type: "excel-intake-agent",
+    description: "excel-intake-agent: write triage results back to Excel M365",
+    prompt: `op: write. channel_config: ${JSON.stringify(channel)}. tickets: ${JSON.stringify(triageResult)}. projectRoot: ${projectRoot}`
+  })
+} else if (channel.type === 'google_sheets') {
+  Agent({ subagent_type: "sheets-intake-agent",
+    description: "sheets-intake-agent: write triage results back to Google Sheets",
+    prompt: `op: write. channel_config: ${JSON.stringify(channel)}. tickets: ${JSON.stringify(triageResult)}. projectRoot: ${projectRoot}`
+  })
+} else if (channel.type === 'browser') {
+  // browser channels are read-only — no write-back (public URLs, no auth)
+  console.log(`ℹ️  Browser channel "${channel.name}" is read-only — skipping write-back.`);
+} else {
+  await writeback(channel, triageResult, projectRoot);       // lib/intake/writeback.cjs (csv + non-agent path)
+}
+```
+
+**Non-CC adapters** — all types via `lib/intake/writeback.cjs`:
+```js
+await writeback(channel, triageResult, projectRoot);
+```
+
 ```js
-await writeback(channel, triageResult, projectRoot);         // lib/intake/writeback.cjs
 const reportPath = generateTriageReport(channel, triageResult, projectRoot);  // lib/intake/report.cjs
 ```
 
@@ -199,6 +496,7 @@ Write-back failure → warn (non-fatal), report is still generated.
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 
  Channel: {channel.name}
+ Manifest: .viepilot/intake/{channelId}-manifest.json  (shown only when manifest was used)
  Accepted: {N} → {request IDs}
  Declined: {N}
  Report: .viepilot/intake/TRIAGE-{timestamp}.md
@@ -226,7 +524,8 @@ options:
 - [ ] Tickets classified as BUG / ENH / UNCLEAR
 - [ ] AUQ multi-select triage completed with accept/decline per ticket
 - [ ] Decline reasons collected and attached
-- [ ] Accepted tickets auto-create `.viepilot/requests/` files
+- [ ] Accepted tickets auto-create `.viepilot/requests/` files with `## Intake Source` block for write-back traceability
+- [ ] Manifest Lifecycle (ENH-095): manifest checked/analyzed before Step 4; column_map overridden from manifest when not explicitly set; `--analyze` flag forces re-analysis
 - [ ] Write-back updates source (non-fatal on failure)
 - [ ] TRIAGE session report generated at `.viepilot/intake/TRIAGE-{timestamp}.md`
 </success_criteria>

package/skills/vp-pause/SKILL.md

@@ -53,7 +53,7 @@ Silent if command unavailable or errors.
 </persona_context>
 
 
-<cursor_skill_adapter>
+<adapter id="claude-code">
 ## A. Skill Invocation
 - Skill được gọi khi user mention `vp-pause`, `/vp-pause`, "pause", "dừng", "tạm nghỉ"
 - Treat all user text after the skill mention as `{{VP_ARGS}}`
@@ -62,8 +62,57 @@ Silent if command unavailable or errors.
 Prompt user conversationally to gather state info.
 
 ## C. Tool Usage
-Use Cursor tools: `Shell`, `ReadFile`, `Glob`, `rg`, `ApplyPatch`, `WebSearch`, `WebFetch`, `Subagent`
-</cursor_skill_adapter>
+Use Claude Code tools: `Bash` (shell), `Read` (file), `Edit` + `Write` (file write/patch),
+`Grep` (search), `Glob` (file patterns), `LS`, `WebSearch`, `WebFetch`,
+`Agent` (spawn subagent — multi-level nesting supported)
+Interactive: `AskUserQuestion` (deferred — preload via ToolSearch before first call)
+</adapter>
+
+<adapter id="cursor-agent">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+
+## C. Tool Usage
+Use Cursor tools: `run_terminal_cmd` (shell), `read_file` (read), `edit_file` (write/edit),
+`grep_search` (search), `web_search`, `codebase_search`, `list_dir`, `file_search`
+Interactive: text list fallback (AskQuestion available in Plan Mode only; Agent Mode = text)
+Subagent: `/multitask` (user command, single-level only — not a callable tool)
+MCP limit: 40 tools
+</adapter>
+
+<adapter id="antigravity">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+Skill discovery: LLM-driven (automatic, no slash command needed).
+
+## C. Tool Usage
+Use Antigravity tools: `shell` (cmd), `file_read`, `file_write`, MCP plugins
+Interactive: text fallback (TUI-based; no formal AskUserQuestion)
+Skill path: `.agents/skills/<skill>/SKILL.md` (project) or `~/.gemini/antigravity/skills/` (global)
+Note: Gemini CLI deprecated June 18, 2026 — use Antigravity CLI.
+</adapter>
+
+<adapter id="codex">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+
+## C. Tool Usage
+Use Codex tools: `container.exec` (sandboxed shell), `apply_patch` (file write), `web_search`
+Interactive: text fallback (TUI Tab/Enter injection)
+Config: `~/.codex/config.toml`
+</adapter>
+
+<adapter id="copilot">
+## A. Skill Invocation
+Same trigger keywords as claude-code adapter.
+Discovery: User-driven (`@agent-name` in GitHub Copilot Chat).
+
+## C. Tool Usage
+Use Copilot tools: `runCommands` (shell), `read`/`readfile` (read), `edit`/`editFiles` (write),
+`code_search`, `find_references`
+Interactive: `askQuestions` (main agent only — NOT available in subagents; VS Code issue #293745)
+Skill path: `.github/agents/<name>.agent.md`
+</adapter>
 <scope_policy>
 ## ViePilot Namespace Guard (BUG-004)
 - Default mode: only use and reference `vp-*` skills in ViePilot workflows.