agent-framework-lib 0.8.1.post4__tar.gz → 0.8.1.post6__tar.gz

{agent_framework_lib-0.8.1.post4/agent_framework_lib.egg-info → agent_framework_lib-0.8.1.post6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agent-framework-lib
-Version: 0.8.1.post4
+Version: 0.8.1.post6
 Summary: A comprehensive Python framework for building and serving conversational AI agents with FastAPI
 Author-email: Sebastian Pavel <sebastian@cinco.ai>, Elliott Girard <elliott.girard@icloud.com>
 Maintainer-email: Sebastian Pavel <sebastian@cinco.ai>

{agent_framework_lib-0.8.1.post4 → agent_framework_lib-0.8.1.post6}/agent_framework/skills/discovery_prompt.py

@@ -28,7 +28,15 @@ create files, render diagrams, etc.). You do NOT call Python tool functions dire
 - `load_skill(skill_name)` - Load a skill to get its instructions and tools
 - `unload_skill(skill_name)` - Unload a skill when no longer needed
 
-## ⚠️ CRITICAL RULE: ALWAYS LOAD SKILLS FIRST - NO EXCEPTIONS
+## ⚠️ CRITICAL RULES — NO EXCEPTIONS
+
+### Rule 1: DISCOVER before declining
+
+**BEFORE saying you cannot do something, you MUST call `list_skills()` first.**
+If a skill exists that could fulfill the request, load it and use it.
+**NEVER tell the user you lack a capability without checking `list_skills()` first.**
+
+### Rule 2: LOAD before using
 
 **You MUST call `load_skill("skill_name")` BEFORE using ANY skill capability.**
 **You MUST NOT invent or assume capabilities that are not explicitly provided.**
@@ -81,113 +89,43 @@ field — use it to give the user a clickable download link.
 ### ✅ CORRECT WORKFLOW (MANDATORY):
 
 ```
-User asks for a chart/diagram/table/etc.
+User asks for ANYTHING that might require a skill (file generation, charts,
+diagrams, tables, PDFs, images, data export, etc.)
+↓
+1. Call list_skills() to discover available skills
+↓
+2. Identify the skill that matches the request
 ↓
-1. Call load_skill("skill_name") FIRST
+3. Call load_skill("skill_name") to get instructions
 ↓
-2. READ the instructions returned by load_skill
+4. READ the instructions returned by load_skill
 ↓
-3. Construct the shell command as described
+5. Construct the shell command as described
 ↓
-4. Execute via shell_exec
+6. Execute via shell_exec
 ```
 
-### Example - Creating a Chart:
+### Example:
 
 **WRONG (will fail):**
 ```
-User: "Create a chart"
-→ Directly write ```chart JSON ❌
-→ Or run a script without loading the skill ❌
-Result: Error or broken output!
+User: "Create a diagram of our architecture"
+→ Write raw Mermaid code in the response ❌
+→ Or say "I can't create diagrams" ❌
 ```
 
 **CORRECT:**
 ```
-User: "Create a chart"
-→ Call load_skill("chart") ✅
+User: "Create a diagram of our architecture"
+→ Call list_skills() — discover mermaid skill ✅
+→ Call load_skill("mermaid") ✅
 → Read the instructions returned
 → Build the shell command as instructed
 → Execute via shell_exec ✅
 ```
 
----
-
-## 📊 Charts & Graphs
-
-**NEVER output raw Chart.js JSON or data descriptions.**
-**ALWAYS load the chart skill first.**
-
-**Mandatory workflow:**
-1. **FIRST:** Call `load_skill("chart")` - this is REQUIRED
-2. Read the instructions returned - they tell you the exact format
-3. Construct the shell command as described in the instructions
-4. Execute via `shell_exec`
-
-**If you skip load_skill("chart"), your chart WILL NOT WORK.**
-
----
-
-## 🔀 Diagrams (Mermaid)
-
-**NEVER output raw Mermaid code or text descriptions.**
-**ALWAYS load the mermaid skill first.**
-
-**Mandatory workflow:**
-1. **FIRST:** Call `load_skill("mermaid")` - this is REQUIRED
-2. Read the instructions returned - they tell you the exact syntax
-3. Construct the shell command as described in the instructions
-4. Execute via `shell_exec`
-
-**If you skip load_skill("mermaid"), your diagram WILL NOT WORK.**
-
----
-
-## 📋 Tabular Data
-
-**NEVER output raw markdown tables or JSON arrays.**
-**ALWAYS load the table skill first.**
-
-**Mandatory workflow:**
-1. **FIRST:** Call `load_skill("table")` - this is REQUIRED
-2. Read the instructions returned - they tell you the exact format
-3. Follow those instructions - do not invent your own format
-4. Use the tabledata format as instructed
-
-**If you skip load_skill("table"), your table WILL NOT RENDER correctly.**
-
----
-
-## 🚨 Options Blocks
-
-**ALWAYS end your responses with clickable options** using the ```optionsblock format.
-
-**Quick format:**
-```optionsblock
-{
-  "question": "What would you like to do next?",
-  "options": [
-    {"text": "✅ Option 1", "value": "option1"},
-    {"text": "🔄 Option 2", "value": "option2"},
-    {"text": "❓ Something else", "value": "other"}
-  ]
-}
-```
-
-For detailed instructions, call `load_skill("optionsblock")`.
-
----
-
-## 🖼️ Displaying Images
-
-**NEVER use markdown syntax** `![alt](url)` to display images.
-
-**For web URLs (http/https), use JSON format directly:**
-```json
-{"image": {"url": "https://example.com/image.png", "alt": "Description", "caption": "Optional caption"}}
-```
-
-For detailed instructions, call `load_skill("image_display")`.
+This applies to ALL skill-based capabilities: charts, diagrams, tables, PDFs,
+Excel files, images, forms, etc. **Always discover first, then load, then use.**
 
 ---
 
@@ -204,12 +142,11 @@ For detailed instructions, call `load_skill("image_display")`.
 - ❌ Use skills that haven't been loaded via load_skill()
 - ❌ Invent new output formats or command parameters
 
-**If you're unsure about a capability:**
-1. Call `list_skills()` to see what's available
-2. Call `load_skill()` to get exact instructions
-3. Follow those instructions - do not improvise
+**Before telling the user something is not possible:**
+1. Call `list_skills()` to check available capabilities
+2. If a matching skill exists, load it and use it
+3. Only if NO skill matches, tell the user it's not available
 
-**If a capability doesn't exist, tell the user it's not available.**
 **Do NOT promise or attempt things you cannot do.**
 
 **ALWAYS RESPOND AS PRECISE AS POSSIBLE TO THE REQUEST OF THE USER**

{agent_framework_lib-0.8.1.post4 → agent_framework_lib-0.8.1.post6}/agent_framework/workspace/artefacts.py

@@ -5,6 +5,7 @@ Provides factory function to create artefact tool definitions
 """
 
 import logging
+from pathlib import Path
 from typing import Any
 
 import httpx
@@ -17,6 +18,38 @@ logger = logging.getLogger(__name__)
 _DEFAULT_MAX_DOWNLOAD_SIZE = 50 * 1024 * 1024  # 50 MB
 
 
+def _save_binary_to_temp(
+    data: bytes,
+    artefact_id: str,
+    file_name: str,
+    base_dir: str = "/tmp/workspace_artefacts",
+) -> Path:
+    """Save binary data to a temp directory and return the file path.
+
+    Args:
+        data: Raw file bytes.
+        artefact_id: Used as subdirectory name.
+        file_name: Original file name with extension.
+        base_dir: Base temp directory path.
+
+    Returns:
+        Absolute path to the saved file.
+
+    Raises:
+        OSError: If directory creation or file write fails.
+    """
+    dir_path = Path(base_dir) / artefact_id
+    file_path = dir_path / file_name
+    try:
+        dir_path.mkdir(parents=True, exist_ok=True)
+        file_path.write_bytes(data)
+    except OSError as exc:
+        raise OSError(
+            f"Failed to save '{file_name}' to temp directory: {exc}"
+        ) from exc
+    return file_path.resolve()
+
+
 def create_artefact_tools(
     client: WorkspaceClient,
     workspace_id: str,
@@ -108,9 +141,12 @@ def create_artefact_tools(
         try:
             return data.decode("utf-8")
         except UnicodeDecodeError:
+            saved_path = _save_binary_to_temp(data, target.id, target.file_name)
+            logger.info("Binary file saved to %s", saved_path)
             return (
                 f"[Binary file: {target.file_name}, "
-                f"{len(data)} bytes, type: {target.type}]"
+                f"{len(data)} bytes, saved to: {saved_path}]\n"
+                f"Use read_file with the path above to extract content."
             )
 
     async def upload_workspace_artefact(
@@ -153,7 +189,9 @@ def create_artefact_tools(
             "name": "download_workspace_artefact",
             "description": (
                 "Download a workspace artefact by its ID. "
-                "Returns the file content as text, or a description for binary files."
+                "Returns text content directly, or saves binary files locally "
+                "and returns the file path. Use the file skill's read_file "
+                "with the returned path to extract content from binary files."
             ),
             "fn": download_workspace_artefact,
         },

{agent_framework_lib-0.8.1.post4 → agent_framework_lib-0.8.1.post6/agent_framework_lib.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agent-framework-lib
-Version: 0.8.1.post4
+Version: 0.8.1.post6
 Summary: A comprehensive Python framework for building and serving conversational AI agents with FastAPI
 Author-email: Sebastian Pavel <sebastian@cinco.ai>, Elliott Girard <elliott.girard@icloud.com>
 Maintainer-email: Sebastian Pavel <sebastian@cinco.ai>

{agent_framework_lib-0.8.1.post4 → agent_framework_lib-0.8.1.post6}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "agent-framework-lib"
-version = "0.8.1post4"
+version = "0.8.1post6"
 description = "A comprehensive Python framework for building and serving conversational AI agents with FastAPI"
 readme = {file = "README.md", content-type = "text/markdown"}
 license = {text = "MIT"}