@buda-ai/connector 0.1.0 → 0.1.2

package/agent-templates/system-skills/buda-automations/scripts/create-automation.sh

@@ -0,0 +1,226 @@
+#!/usr/bin/env bash
+# Create a new scheduled automation in the current agent's workspace.
+#
+# Usage:
+#   bash create-automation.sh \
+#     --name "Task Name" \
+#     --prompt "What the agent should do" \
+#     --schedule-type interval \
+#     --interval-minutes 60
+#
+#   bash create-automation.sh \
+#     --name "Task Name" \
+#     --prompt "What the agent should do" \
+#     --schedule-type cron \
+#     --cron "0 9 * * 1-5"
+#
+#   bash create-automation.sh \
+#     --name "Task Name" \
+#     --prompt "What the agent should do" \
+#     --schedule-type once \
+#     --scheduled-at "2026-03-25T10:00:00.000Z"
+#
+# Options:
+#   --name              Required. Automation display name.
+#   --prompt            Required. The prompt the agent receives when triggered.
+#   --schedule-type     Required. One of: interval, cron, once.
+#   --interval-minutes  Required for interval type.
+#   --cron              Required for cron type. Cron expression string.
+#   --scheduled-at      Required for once type. ISO 8601 datetime.
+#   --max-runs          Optional. Maximum number of executions.
+#   --disabled          Optional flag. Create in disabled state.
+#   --delivery-mode     Optional. One of: session, notify (default: session).
+#   --channel-id        Optional. Target channel ID for notify mode.
+#   --timezone          Optional. IANA timezone for cron schedules (e.g. Asia/Shanghai, America/New_York).
+#                       The cron time is interpreted in this local timezone.
+#
+# Requires: AGENT_KEY, NEXT_PUBLIC_APP_URL (auto-injected in Buda sandbox)
+
+set -euo pipefail
+
+NAME=""
+PROMPT=""
+SCHEDULE_TYPE=""
+INTERVAL_MINUTES=""
+CRON_EXPR=""
+SCHEDULED_AT=""
+MAX_RUNS=""
+IS_ENABLED="true"
+DELIVERY_MODE="session"
+CHANNEL_ID=""
+TIMEZONE=""
+POSITIONAL_ARGS=()
+
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    --name)             NAME="$2"; shift 2 ;;
+    --prompt)           PROMPT="$2"; shift 2 ;;
+    --schedule-type)    SCHEDULE_TYPE="$2"; shift 2 ;;
+    --interval-minutes) INTERVAL_MINUTES="$2"; shift 2 ;;
+    --cron)             CRON_EXPR="$2"; shift 2 ;;
+    --scheduled-at)     SCHEDULED_AT="$2"; shift 2 ;;
+    --max-runs)         MAX_RUNS="$2"; shift 2 ;;
+    --delivery-mode)    DELIVERY_MODE="$2"; shift 2 ;;
+    --channel-id)       CHANNEL_ID="$2"; shift 2 ;;
+    --timezone)         TIMEZONE="$2"; shift 2 ;;
+    --disabled)         IS_ENABLED="false"; shift ;;
+    # Be tolerant: free-text arguments are treated as prompt fallback.
+    *) POSITIONAL_ARGS+=("$1"); shift ;;
+  esac
+done
+
+# Fallback for accidental natural-language invocation:
+#   bash create-automation.sh remind me tomorrow 9am
+if [[ -z "${PROMPT}" && ${#POSITIONAL_ARGS[@]} -gt 0 ]]; then
+  PROMPT="${POSITIONAL_ARGS[*]}"
+fi
+if [[ -z "${NAME}" && -n "${PROMPT}" ]]; then
+  NAME="Automation $(date +%s)"
+fi
+
+# Validate required fields
+if [[ -z "${NAME}" ]]; then
+  echo "ERROR: --name is required." >&2; exit 1
+fi
+if [[ -z "${PROMPT}" ]]; then
+  echo "ERROR: --prompt is required." >&2; exit 1
+fi
+if [[ -z "${SCHEDULE_TYPE}" ]]; then
+  echo "ERROR: --schedule-type is required (interval, cron, or once)." >&2
+  echo "Hint: pass schedule flags explicitly, e.g. --schedule-type cron --cron \"0 9 * * *\"" >&2
+  exit 1
+fi
+if [[ -z "${AGENT_KEY:-}" ]]; then
+  echo "ERROR: AGENT_KEY environment variable is not set." >&2; exit 1
+fi
+if [[ -z "${NEXT_PUBLIC_APP_URL:-}" ]]; then
+  echo "ERROR: NEXT_PUBLIC_APP_URL environment variable is not set." >&2; exit 1
+fi
+
+# Validate schedule-type-specific fields
+case "${SCHEDULE_TYPE}" in
+  interval)
+    if [[ -z "${INTERVAL_MINUTES}" ]]; then
+      echo "ERROR: --interval-minutes is required for interval schedule type." >&2; exit 1
+    fi
+    ;;
+  cron)
+    if [[ -z "${CRON_EXPR}" ]]; then
+      echo "ERROR: --cron is required for cron schedule type." >&2; exit 1
+    fi
+    if [[ -z "${TIMEZONE}" ]]; then
+      echo "ERROR: --timezone is required for cron schedule type." >&2
+      echo "Hint: pass the user's IANA timezone, e.g. --timezone \"Asia/Shanghai\"" >&2
+      exit 1
+    fi
+    ;;
+  once)
+    if [[ -z "${SCHEDULED_AT}" ]]; then
+      echo "ERROR: --scheduled-at is required for once schedule type." >&2; exit 1
+    fi
+    ;;
+  *)
+    echo "ERROR: Invalid schedule type '${SCHEDULE_TYPE}'. Must be interval, cron, or once." >&2
+    exit 1
+    ;;
+esac
+
+# Validate delivery fields
+if [[ "${DELIVERY_MODE}" != "session" && "${DELIVERY_MODE}" != "notify" ]]; then
+  echo "ERROR: --delivery-mode must be 'session' or 'notify'." >&2; exit 1
+fi
+
+# Channel-originated automation must deliver back to the same channel.
+# BUDA_SOURCE_CHANNEL_ID is injected by Buda when chat originates from a channel.
+# BUDA_SESSION_ID is the current session ID (maps to a specific group/DM via chatSessions).
+if [[ -n "${BUDA_SOURCE_CHANNEL_ID:-}" ]]; then
+  DELIVERY_MODE="notify"
+  CHANNEL_ID="${BUDA_SOURCE_CHANNEL_ID}"
+fi
+
+SESSION_ID="${BUDA_SESSION_ID:-}"
+
+if [[ "${DELIVERY_MODE}" == "notify" && -z "${CHANNEL_ID}" ]]; then
+  echo "ERROR: --channel-id is required when --delivery-mode notify is used." >&2; exit 1
+fi
+
+# Build JSON payload using jq for proper escaping
+JSON_PAYLOAD=$(jq -n \
+  --arg name "${NAME}" \
+  --arg prompt "${PROMPT}" \
+  --arg scheduleType "${SCHEDULE_TYPE}" \
+  --arg cronExpression "${CRON_EXPR}" \
+  --arg intervalMinutes "${INTERVAL_MINUTES}" \
+  --arg scheduledAt "${SCHEDULED_AT}" \
+  --arg maxRuns "${MAX_RUNS}" \
+  --arg deliveryMode "${DELIVERY_MODE}" \
+  --arg channelId "${CHANNEL_ID}" \
+  --arg sessionId "${SESSION_ID}" \
+  --arg timezone "${TIMEZONE}" \
+  --argjson isEnabled "${IS_ENABLED}" \
+  '{
+    json: {
+      name: $name,
+      taskTemplate: {
+        title: $name,
+        prompt: $prompt,
+        delivery: {
+          mode: $deliveryMode
+        }
+      },
+      scheduleType: $scheduleType,
+      isEnabled: $isEnabled
+    }
+  }
+  | if $deliveryMode == "notify" then .json.taskTemplate.delivery.channelId = $channelId else . end
+  | if $sessionId != "" then .json.taskTemplate.delivery.sessionId = $sessionId else . end
+  | if $scheduleType == "cron" then .json.cronExpression = $cronExpression else . end
+  | if $scheduleType == "interval" then .json.intervalMinutes = ($intervalMinutes | tonumber) else . end
+  | if $scheduleType == "once" then .json.scheduledAt = $scheduledAt else . end
+  | if $maxRuns != "" then .json.maxRuns = ($maxRuns | tonumber) else . end
+  | if $timezone != "" then .json.timezone = $timezone else . end
+  ')
+
+RESPONSE=$(curl -s \
+  "${NEXT_PUBLIC_APP_URL}/api/rpc/agent/automations/create" \
+  -X POST \
+  -H "Authorization: Bearer ${AGENT_KEY}" \
+  -H "Content-Type: application/json" \
+  -w '\n%{http_code}' \
+  -d "${JSON_PAYLOAD}" 2>/dev/null) || {
+  echo "ERROR: Failed to connect to API." >&2
+  exit 1
+}
+
+HTTP_CODE=$(echo "${RESPONSE}" | tail -1)
+BODY=$(echo "${RESPONSE}" | sed '$d')
+
+if [[ "${HTTP_CODE}" -ge 400 ]]; then
+  # Extract the human-readable error message from oRPC JSON error response
+  ERROR_MSG=$(echo "${BODY}" | jq -r '
+    (.json.message // "") as $msg |
+    (.json.code // "") as $code |
+    (if ($msg | type) == "string" and ($msg | startswith("[")) then
+      (try ($msg | fromjson | map((.path | join(".")) + ": " + .message) | join("; ")) catch $msg)
+    else
+      $msg
+    end) as $parsed_msg |
+    if $code != "" and $parsed_msg != "" then
+      "[\($code)] \($parsed_msg)"
+    elif $parsed_msg != "" then
+      $parsed_msg
+    else
+      empty
+    end
+  ' 2>/dev/null) || ERROR_MSG=""
+
+  if [[ -n "${ERROR_MSG}" ]]; then
+    echo "ERROR: ${ERROR_MSG}" >&2
+  else
+    echo "ERROR: API call failed (HTTP ${HTTP_CODE})." >&2
+    echo "${BODY}" >&2
+  fi
+  exit 1
+fi
+
+echo "${BODY}" | jq -r '.json'

package/agent-templates/system-skills/buda-automations/scripts/delete-automation.sh

@@ -0,0 +1,73 @@
+#!/usr/bin/env bash
+# Delete a scheduled automation by ID.
+#
+# Usage: bash delete-automation.sh --id "sch_xxxxxxxxxxxxxxxx"
+#
+# Requires: AGENT_KEY, NEXT_PUBLIC_APP_URL (auto-injected in Buda sandbox)
+
+set -euo pipefail
+
+AUTOMATION_ID=""
+
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    --id) AUTOMATION_ID="$2"; shift 2 ;;
+    *) echo "Unknown option: $1" >&2; exit 1 ;;
+  esac
+done
+
+if [[ -z "${AUTOMATION_ID}" ]]; then
+  echo "ERROR: --id is required." >&2; exit 1
+fi
+if [[ -z "${AGENT_KEY:-}" ]]; then
+  echo "ERROR: AGENT_KEY environment variable is not set." >&2; exit 1
+fi
+if [[ -z "${NEXT_PUBLIC_APP_URL:-}" ]]; then
+  echo "ERROR: NEXT_PUBLIC_APP_URL environment variable is not set." >&2; exit 1
+fi
+
+PAYLOAD=$(jq -n --arg id "${AUTOMATION_ID}" '{ json: { id: $id } }')
+
+RESPONSE=$(curl -s \
+  "${NEXT_PUBLIC_APP_URL}/api/rpc/agent/automations/delete" \
+  -X POST \
+  -H "Authorization: Bearer ${AGENT_KEY}" \
+  -H "Content-Type: application/json" \
+  -w '\n%{http_code}' \
+  -d "${PAYLOAD}" 2>/dev/null) || {
+  echo "ERROR: Failed to connect to API." >&2
+  exit 1
+}
+
+HTTP_CODE=$(echo "${RESPONSE}" | tail -1)
+BODY=$(echo "${RESPONSE}" | sed '$d')
+
+if [[ "${HTTP_CODE}" -ge 400 ]]; then
+  # Extract the human-readable error message from oRPC JSON error response
+  ERROR_MSG=$(echo "${BODY}" | jq -r '
+    (.json.message // "") as $msg |
+    (.json.code // "") as $code |
+    (if ($msg | type) == "string" and ($msg | startswith("[")) then
+      (try ($msg | fromjson | map((.path | join(".")) + ": " + .message) | join("; ")) catch $msg)
+    else
+      $msg
+    end) as $parsed_msg |
+    if $code != "" and $parsed_msg != "" then
+      "[\($code)] \($parsed_msg)"
+    elif $parsed_msg != "" then
+      $parsed_msg
+    else
+      empty
+    end
+  ' 2>/dev/null) || ERROR_MSG=""
+
+  if [[ -n "${ERROR_MSG}" ]]; then
+    echo "ERROR: ${ERROR_MSG}" >&2
+  else
+    echo "ERROR: API call failed (HTTP ${HTTP_CODE})." >&2
+    echo "${BODY}" >&2
+  fi
+  exit 1
+fi
+
+echo "${BODY}" | jq -r '.json'

package/agent-templates/system-skills/buda-automations/scripts/list-automations.sh

@@ -0,0 +1,77 @@
+#!/usr/bin/env bash
+# List all automations in the current agent's workspace.
+# Usage: bash list-automations.sh [--limit N] [--offset N]
+#
+# Requires: AGENT_KEY, NEXT_PUBLIC_APP_URL (auto-injected in Buda sandbox)
+
+set -euo pipefail
+
+LIMIT=20
+OFFSET=0
+
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    --limit)  LIMIT="$2"; shift 2 ;;
+    --offset) OFFSET="$2"; shift 2 ;;
+    *) echo "Unknown option: $1" >&2; exit 1 ;;
+  esac
+done
+
+if [[ -z "${AGENT_KEY:-}" ]]; then
+  echo "ERROR: AGENT_KEY environment variable is not set." >&2
+  exit 1
+fi
+if [[ -z "${NEXT_PUBLIC_APP_URL:-}" ]]; then
+  echo "ERROR: NEXT_PUBLIC_APP_URL environment variable is not set." >&2
+  exit 1
+fi
+
+PAYLOAD=$(cat <<EOF
+{"json":{"limit":${LIMIT},"offset":${OFFSET}}}
+EOF
+)
+
+RESPONSE=$(curl -s \
+  "${NEXT_PUBLIC_APP_URL}/api/rpc/agent/automations/list" \
+  -X POST \
+  -H "Authorization: Bearer ${AGENT_KEY}" \
+  -H "Content-Type: application/json" \
+  -w '\n%{http_code}' \
+  -d "${PAYLOAD}" 2>/dev/null) || {
+  echo "ERROR: Failed to connect to API." >&2
+  exit 1
+}
+
+HTTP_CODE=$(echo "${RESPONSE}" | tail -1)
+BODY=$(echo "${RESPONSE}" | sed '$d')
+
+if [[ "${HTTP_CODE}" -ge 400 ]]; then
+  # Extract the human-readable error message from oRPC JSON error response
+  ERROR_MSG=$(echo "${BODY}" | jq -r '
+    (.json.message // "") as $msg |
+    (.json.code // "") as $code |
+    (if ($msg | type) == "string" and ($msg | startswith("[")) then
+      (try ($msg | fromjson | map((.path | join(".")) + ": " + .message) | join("; ")) catch $msg)
+    else
+      $msg
+    end) as $parsed_msg |
+    if $code != "" and $parsed_msg != "" then
+      "[\($code)] \($parsed_msg)"
+    elif $parsed_msg != "" then
+      $parsed_msg
+    else
+      empty
+    end
+  ' 2>/dev/null) || ERROR_MSG=""
+
+  if [[ -n "${ERROR_MSG}" ]]; then
+    echo "ERROR: ${ERROR_MSG}" >&2
+  else
+    echo "ERROR: API call failed (HTTP ${HTTP_CODE})." >&2
+    echo "${BODY}" >&2
+  fi
+  exit 1
+fi
+
+# Extract the result data
+echo "${BODY}" | jq -r '.json'

package/agent-templates/system-skills/buda-far/SKILL.md

@@ -0,0 +1,221 @@
+---
+name: far
+description: File-Augmented Retrieval — generate persistent .meta sidecar files for PDFs, images, spreadsheets, videos and more, making every binary file readable to AI coding agents.
+allowed-tools: Read, Write, Edit, Bash, Glob
+user-invocable: true
+---
+
+# 📄 /far - File-Augmented Retrieval
+
+> **"Making Every File Readable to Coding Agents via Persistent .meta Sidecars"**
+
+FAR is a file augmentation protocol that generates persistent `.meta` sidecar files for binary documents (PDF, DOCX, XLSX, PPTX, Images, etc.). This allows AI coding agents (like OpenClaw, Cursor, GitHub Copilot) to "read" non-text files directly from the file system without requiring external RAG infrastructure.
+
+**Current Version:** 1.0.0
+**Author:** Kelly Peilin Chan
+
+---
+
+## 🚀 The Problem: AI "Blindness"
+
+AI agents operating in a repository can read code (`.py`, `.js`, `.md`), but they are blind to 30-40% of critical context stored in binary formats:
+- **Product Specs**: `requirements.docx`
+- **Design Mocks**: `architecture.png`
+- **Financial Data**: `budget.xlsx`
+- **Contracts**: `agreement.pdf`
+
+When an agent encounters `budget.xlsx`, it sees opaque bytes. It cannot reason about the content.
+
+## 💡 The Solution: Persistent Sidecars
+
+FAR solves this by generating a human-readable and machine-readable `.meta` file next to every binary file.
+
+**Example:**
+```
+project/
+├── budget.xlsx         (Binary, opaque)
+└── budget.xlsx.meta    (Markdown, readable)
+```
+
+The agent simply reads `budget.xlsx.meta` to understand the spreadsheet. No vector database, no API calls, no runtime overhead.
+
+---
+
+## ✨ Features
+
+### 1. Broad Format Support
+FAR extracts text and structure from a wide range of formats:
+*   **📄 PDF** (`.pdf`): Full text extraction with layout preservation. **OCR fallback** for scanned PDFs. **Embedded image extraction** (pdfimages + OCR/Vision).
+*   **📝 Word** (`.docx`, `.doc`): Text extraction.
+*   **📊 Excel** (`.xlsx`): Sheet data converted to Markdown tables.
+*   **📽️ PowerPoint** (`.pptx`): Slide text extraction.
+*   **🖼️ Images** (`.png`, `.jpg`, `.jpeg`, `.gif`, `.bmp`, `.webp`): **OCR** (Tesseract) + **Apple Vision (on-device, macOS)** + **AI Vision** (GPT-4o) if configured.
+*   **🎬 Media** (`.mp4`, `.mov`, `.mp3`, `.wav`, `.m4a`, `.flac`): Technical metadata (FFprobe) + **Apple Vision video-frame analysis (macOS)** + **AI Transcription** through Buda's `whisper` tool when available.
+*   **📋 CSV** (`.csv`): Data rendered as Markdown tables (up to 100 rows).
+*   **📓 Jupyter Notebook** (`.ipynb`): Markdown cells, code cells, and outputs.
+*   **📚 EPUB** (`.epub`): Full text extracted from all chapters in spine order.
+*   **📦 Tar** (`.tar`, `.tar.gz`, `.tgz`, `.bz2`, `.xz`): File listing with sizes.
+*   **📧 Email** (`.eml`, `.msg`): Headers, body text, and attachment list.
+*   **📝 RTF** (`.rtf`): Plain text extraction via control word stripping.
+*   **🗄️ SQLite** (`.db`, `.sqlite`, `.sqlite3`): Table schemas + latest 20 rows per table as Markdown tables.
+*   **📊 Parquet** (`.parquet`): Schema and row count via `pyarrow` (optional). *[Metadata only]*
+*   **🎨 Design** (`.fig`, `.sketch`, `.xd`): File size and page/canvas count. *[Metadata only]*
+*   **💻 Code/Text** (`.py`, `.js`, `.ts`, `.go`, `.rs`, `.md`, `.json`, ...): Direct content mirroring.
+
+### 2. Intelligent Caching (Incremental Build)
+FAR is designed for speed. It uses a **two-layer caching mechanism**:
+1.  **Fast Check (mtime & size)**: If the file hasn't been modified, it skips processing instantly (0.003s).
+2.  **Content Check (SHA256)**: If mtime changed, it calculates the file hash. If the content is identical, it updates the timestamp but skips re-extraction.
+
+For directory summaries, FAR also avoids noisy rewrites: if `.dir.meta` summary content is unchanged, it keeps the file untouched (no `extracted_at` churn).
+
+### 3. Directory Summaries (`.dir.meta`)
+FAR generates a `.dir.meta` file in every directory, providing a high-level summary of all files within. This allows agents to "browse" a folder and understand its contents without reading every single file.
+
+### 4. Git LFS Support
+FAR automatically handles Git LFS pointer files. It will attempt to pull the real content before processing, ensuring you don't get empty metadata for LFS-tracked files.
+
+---
+
+## 🛠️ Usage
+
+### Installation
+The skill is pre-installed in the OpenClaw workspace:
+```bash
+~/.openclaw/workspace/.agents/skills/far/far_gen.py
+```
+A symlink is available as `far`:
+```bash
+far [directory_or_file]
+```
+
+### Commands
+
+**Scan current directory (Recursive):**
+```bash
+far
+```
+
+**Scan specific directory:**
+```bash
+far ~/Documents/projects/files
+```
+
+**Process single file:**
+```bash
+far report.pdf
+```
+
+**Force regeneration (Ignore cache):**
+```bash
+far . --force
+```
+
+### Configuration (AI Features)
+In Buda sandboxes, audio/video transcription uses the Buda-native `whisper` tool through the auto-injected `AGENT_KEY` and `NEXT_PUBLIC_APP_URL` environment variables. No user OpenAI key is required inside the sandbox for transcription.
+
+For standalone FAR usage outside Buda, or for AI Vision image description fallback, create a `.env` file in the skill directory or your home folder `~/.far.env`.
+Copy `.env.example` as a template:
+
+```bash
+OPENAI_API_KEY=sk-your-key-here
+# Optional
+OPENAI_BASE_URL=https://api.openai.com/v1
+# Optional (macOS default: enabled)
+FAR_USE_APPLE_VISION=1
+# Optional (macOS default: enabled)
+FAR_USE_MACOS_METADATA=1
+# Optional tuning
+FAR_APPLE_VISION_TIMEOUT=25
+FAR_APPLE_VISION_MAX_FRAMES=6
+```
+
+If API keys are missing, FAR gracefully falls back to local tools (Tesseract, FFprobe),
+and on macOS it also uses Apple Vision + Spotlight metadata on-device by default.
+
+### Configuration (Ignore)
+Create a `.farignore` file in your project root to exclude files or directories from scanning.
+```gitignore
+# .farignore
+node_modules
+.git
+secrets/
+*.tmp
+```
+
+---
+
+## 🤖 Agent Integration
+
+Add one rule to `AGENTS.md` or your system prompt:
+
+```
+## File Reading with FAR
+
+When navigating a directory, first read `.dir.meta` for a high-level overview
+before opening individual files.
+
+When you encounter a binary file you cannot read directly
+(.png, .pdf, .xlsx, .mp4, .fig, .db, etc.), check for a .meta file
+beside it — it contains the extracted content as Markdown.
+
+Navigation strategy:
+1. Read <dir>/.dir.meta to understand directory structure
+2. Read <file>.meta to access binary file content
+3. Fall back to reading the source file only if no .meta exists
+```
+
+---
+
+## 🏗️ Protocol Specification (v1)
+
+Each `.meta` file follows a strict format with a YAML frontmatter header and Markdown body.
+
+```markdown
+--far_version: 1
+source:
+  sha256: 5a1cc2b8d...
+  mime: application/pdf
+  size: 129509
+  mtime: 1708845210.5
+extract:
+  pipeline: far_gen_v14
+  extracted_at: 2026-02-27T10:00:00Z
+  deterministic: true
+layout:
+  pages: 24
+---
+# filename.pdf
+
+[Extracted Content Here...]
+```
+
+Cache invalidation: a `.meta` is stale if `sha256` or `pipeline` version has changed.
+
+---
+
+## 🤝 Contributing
+
+**Roadmap:**
+- [x] PDF/Word/Excel/PowerPoint Support
+- [x] Incremental Caching (SHA-256 + mtime)
+- [x] Directory Summaries (`.dir.meta`)
+- [x] OCR for Images/Scanned PDFs (Tesseract)
+- [x] Media Metadata (FFprobe) + Transcription (Whisper)
+- [x] AI Vision for Images (GPT-4o)
+- [x] Apple Vision for Images (on-device, macOS)
+- [x] Apple Vision for Video Frames (on-device labels/OCR/faces/barcodes/pose)
+- [x] Apple Vision Feature Print fingerprint (on-device embedding hash)
+- [x] macOS Spotlight metadata enrichment (mdls)
+- [x] PDF Embedded Image Extraction
+- [x] CSV → Markdown Tables
+- [x] Jupyter Notebook Support
+- [x] EPUB, ZIP/TAR, Email, RTF
+- [x] SQLite Schema + Data Preview
+- [x] Parquet Schema, Design File Metadata
+- [x] Pipeline version stale detection
+- [x] Layout frontmatter (pages/sheets/slides)
+- [x] Full glob support in `.farignore`
+- [ ] MCP resource server integration
+- [ ] PII redaction rules
+- [ ] Rust implementation for CI/CD speed