proseforge-agent 0.3.2__tar.gz → 0.4.0__tar.gz

proseforge_agent-0.4.0/PKG-INFO

@@ -0,0 +1,133 @@
+Metadata-Version: 2.4
+Name: proseforge-agent
+Version: 0.4.0
+Summary: General agent runtime with a built-in ProseForge writing workflow pack.
+Author: ProseForge Agent
+License: Apache-2.0
+Requires-Python: <3.13,>=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: pyyaml>=6.0
+Requires-Dist: rich>=13.0
+Requires-Dist: prompt_toolkit>=3.0
+Requires-Dist: tomli>=2.0; python_version < "3.11"
+Provides-Extra: dev
+Requires-Dist: pytest>=7.0; extra == "dev"
+Provides-Extra: ci
+Requires-Dist: pytest>=7.0; extra == "ci"
+Requires-Dist: pytest-rerunfailures>=13.0; extra == "ci"
+Requires-Dist: pytest-repeat>=0.9; extra == "ci"
+Dynamic: license-file
+
+> 🌐 [English](README.md) · [简体中文](README.zh-CN.md)
+
+<p align="center">
+  <img src="assets/banner.png" alt="ProseForge Agent command center" width="100%">
+</p>
+
+<h1 align="center">ProseForge Agent</h1>
+
+<p align="center">
+  <strong>A general agent runtime with a built-in ProseForge writing workflow pack.</strong>
+</p>
+
+<p align="center">
+  <a href="README.zh-CN.md">简体中文</a> ·
+  <a href="https://pypi.org/project/proseforge-agent/">PyPI</a> ·
+  <a href="https://github.com/remacheybn408-boop/ProseForge-Agent/releases">Releases</a> ·
+  <a href="https://github.com/remacheybn408-boop/ProseForge-Agent/pkgs/container/proseforge-agent">GHCR</a>
+</p>
+
+<p align="center">
+  <a href="https://pypi.org/project/proseforge-agent/"><img src="https://img.shields.io/pypi/v/proseforge-agent?style=for-the-badge&label=PyPI" alt="PyPI version"></a>
+  <a href="https://pypi.org/project/proseforge-agent/"><img src="https://img.shields.io/pypi/pyversions/proseforge-agent?style=for-the-badge" alt="Python versions"></a>
+  <a href="https://github.com/remacheybn408-boop/ProseForge-Agent/actions/workflows/ci.yml"><img src="https://img.shields.io/github/actions/workflow/status/remacheybn408-boop/ProseForge-Agent/ci.yml?branch=main&style=for-the-badge&label=CI" alt="CI status"></a>
+  <a href="LICENSE"><img src="https://img.shields.io/badge/License-Apache--2.0-green?style=for-the-badge" alt="Apache-2.0 license"></a>
+</p>
+
+ProseForge Agent is becoming a general personal agent runtime: a terminal-first system for model setup, sessions, tools, approvals, memory, workflow packs, and project automation. The ProseForge writing workflow pack is the first built-in pack, focused on long-form planning, drafting, review, rewrite, export, and memory closeout.
+
+The terminal experience follows modern agent CLI patterns inspired by [openai/codex](https://github.com/openai/codex), while the implementation stays in the Python CLI stack used by projects like [NousResearch/hermes-agent](https://github.com/NousResearch/hermes-agent): `prompt_toolkit` for interaction, `rich` for readable terminal rendering, and Python as the authoritative runtime.
+
+<table>
+<tr><td><b>General runtime</b></td><td>Chat, setup, provider/model switching, sessions, workflow packs, doctor checks, and project binding from one CLI.</td></tr>
+<tr><td><b>Workflow packs</b></td><td>Start with General Chat and ProseForge Novel; extend later into research, coding, docs, automation, and personal assistant packs.</td></tr>
+<tr><td><b>ProseForge writing pack</b></td><td>Project init, chapter drafting, review, rewrite, export, and memory closeout for long-form writing.</td></tr>
+<tr><td><b>Hermes-style terminal stack</b></td><td>Python + <code>prompt_toolkit</code> + <code>rich</code>; no Rust binary is required for ordinary users.</td></tr>
+<tr><td><b>Bring your model</b></td><td>Profiles for OpenAI, Anthropic, Gemini, xAI/Grok, DeepSeek, Qwen, GLM, Mimo, MiniMax, Doubao, plus an offline fake provider.</td></tr>
+</table>
+
+---
+
+## Quick Install
+
+### Python package
+
+```bash
+pip install -U proseforge-agent
+pf-agent --help
+```
+
+### Linux, macOS, WSL2
+
+```bash
+curl -fsSL https://raw.githubusercontent.com/remacheybn408-boop/ProseForge-Agent/main/scripts/install.sh | sh
+```
+
+### Windows PowerShell
+
+```powershell
+iex (irm https://raw.githubusercontent.com/remacheybn408-boop/ProseForge-Agent/main/scripts/install.ps1)
+```
+
+The one-line installer uses PyPI + `uv tool` by default. It installs or upgrades `proseforge-agent`, puts the manager bin directory first on PATH, runs `pf-agent doctor`, then starts `pf-agent setup --quick`.
+
+### Docker
+
+```bash
+docker compose up
+docker compose run --rm pf-agent pf-agent quickstart
+```
+
+The default `docker-compose.yml` uses the published GHCR image:
+`ghcr.io/remacheybn408-boop/proseforge-agent:latest`. It does not build from
+source, so normal users do not need Python, uv, or a local checkout beyond the
+compose file.
+
+### Secrets and env files
+
+`pf-agent setup --quick` runs the guided setup. setup --quick writes provider keys to `.pf-agent/.env`.
+config.yaml does not store raw API keys; it stores provider metadata and key
+references. Environment precedence is:
+
+```text
+shell env > .env.local > .env > .pf-agent/.env > machine defaults
+```
+
+---
+
+## Start
+
+```bash
+pf-agent quickstart
+pf-agent setup --quick
+pf-agent
+pf-agent workflow list
+pf-agent workflow use proseforge
+pf-agent project init --type novel --slug my-novel
+pf-agent doctor
+```
+
+Bare `pf-agent` opens the general Agent Runtime TUI. The legacy stdlib terminal surface remains available with:
+
+```bash
+pf-agent tui --legacy
+```
+
+## Workflow Packs
+
+`general` is the default open-ended runtime workflow. `proseforge` activates the ProseForge writing workflow pack for novel and long-form writing projects. Future packs can add research, coding, document, automation, and personal assistant flows without redefining the core agent runtime.
+
+## License
+
+Apache-2.0. See `LICENSE`.

proseforge_agent-0.4.0/README.md

@@ -0,0 +1,112 @@
+> 🌐 [English](README.md) · [简体中文](README.zh-CN.md)
+
+<p align="center">
+  <img src="assets/banner.png" alt="ProseForge Agent command center" width="100%">
+</p>
+
+<h1 align="center">ProseForge Agent</h1>
+
+<p align="center">
+  <strong>A general agent runtime with a built-in ProseForge writing workflow pack.</strong>
+</p>
+
+<p align="center">
+  <a href="README.zh-CN.md">简体中文</a> ·
+  <a href="https://pypi.org/project/proseforge-agent/">PyPI</a> ·
+  <a href="https://github.com/remacheybn408-boop/ProseForge-Agent/releases">Releases</a> ·
+  <a href="https://github.com/remacheybn408-boop/ProseForge-Agent/pkgs/container/proseforge-agent">GHCR</a>
+</p>
+
+<p align="center">
+  <a href="https://pypi.org/project/proseforge-agent/"><img src="https://img.shields.io/pypi/v/proseforge-agent?style=for-the-badge&label=PyPI" alt="PyPI version"></a>
+  <a href="https://pypi.org/project/proseforge-agent/"><img src="https://img.shields.io/pypi/pyversions/proseforge-agent?style=for-the-badge" alt="Python versions"></a>
+  <a href="https://github.com/remacheybn408-boop/ProseForge-Agent/actions/workflows/ci.yml"><img src="https://img.shields.io/github/actions/workflow/status/remacheybn408-boop/ProseForge-Agent/ci.yml?branch=main&style=for-the-badge&label=CI" alt="CI status"></a>
+  <a href="LICENSE"><img src="https://img.shields.io/badge/License-Apache--2.0-green?style=for-the-badge" alt="Apache-2.0 license"></a>
+</p>
+
+ProseForge Agent is becoming a general personal agent runtime: a terminal-first system for model setup, sessions, tools, approvals, memory, workflow packs, and project automation. The ProseForge writing workflow pack is the first built-in pack, focused on long-form planning, drafting, review, rewrite, export, and memory closeout.
+
+The terminal experience follows modern agent CLI patterns inspired by [openai/codex](https://github.com/openai/codex), while the implementation stays in the Python CLI stack used by projects like [NousResearch/hermes-agent](https://github.com/NousResearch/hermes-agent): `prompt_toolkit` for interaction, `rich` for readable terminal rendering, and Python as the authoritative runtime.
+
+<table>
+<tr><td><b>General runtime</b></td><td>Chat, setup, provider/model switching, sessions, workflow packs, doctor checks, and project binding from one CLI.</td></tr>
+<tr><td><b>Workflow packs</b></td><td>Start with General Chat and ProseForge Novel; extend later into research, coding, docs, automation, and personal assistant packs.</td></tr>
+<tr><td><b>ProseForge writing pack</b></td><td>Project init, chapter drafting, review, rewrite, export, and memory closeout for long-form writing.</td></tr>
+<tr><td><b>Hermes-style terminal stack</b></td><td>Python + <code>prompt_toolkit</code> + <code>rich</code>; no Rust binary is required for ordinary users.</td></tr>
+<tr><td><b>Bring your model</b></td><td>Profiles for OpenAI, Anthropic, Gemini, xAI/Grok, DeepSeek, Qwen, GLM, Mimo, MiniMax, Doubao, plus an offline fake provider.</td></tr>
+</table>
+
+---
+
+## Quick Install
+
+### Python package
+
+```bash
+pip install -U proseforge-agent
+pf-agent --help
+```
+
+### Linux, macOS, WSL2
+
+```bash
+curl -fsSL https://raw.githubusercontent.com/remacheybn408-boop/ProseForge-Agent/main/scripts/install.sh | sh
+```
+
+### Windows PowerShell
+
+```powershell
+iex (irm https://raw.githubusercontent.com/remacheybn408-boop/ProseForge-Agent/main/scripts/install.ps1)
+```
+
+The one-line installer uses PyPI + `uv tool` by default. It installs or upgrades `proseforge-agent`, puts the manager bin directory first on PATH, runs `pf-agent doctor`, then starts `pf-agent setup --quick`.
+
+### Docker
+
+```bash
+docker compose up
+docker compose run --rm pf-agent pf-agent quickstart
+```
+
+The default `docker-compose.yml` uses the published GHCR image:
+`ghcr.io/remacheybn408-boop/proseforge-agent:latest`. It does not build from
+source, so normal users do not need Python, uv, or a local checkout beyond the
+compose file.
+
+### Secrets and env files
+
+`pf-agent setup --quick` runs the guided setup. setup --quick writes provider keys to `.pf-agent/.env`.
+config.yaml does not store raw API keys; it stores provider metadata and key
+references. Environment precedence is:
+
+```text
+shell env > .env.local > .env > .pf-agent/.env > machine defaults
+```
+
+---
+
+## Start
+
+```bash
+pf-agent quickstart
+pf-agent setup --quick
+pf-agent
+pf-agent workflow list
+pf-agent workflow use proseforge
+pf-agent project init --type novel --slug my-novel
+pf-agent doctor
+```
+
+Bare `pf-agent` opens the general Agent Runtime TUI. The legacy stdlib terminal surface remains available with:
+
+```bash
+pf-agent tui --legacy
+```
+
+## Workflow Packs
+
+`general` is the default open-ended runtime workflow. `proseforge` activates the ProseForge writing workflow pack for novel and long-form writing projects. Future packs can add research, coding, document, automation, and personal assistant flows without redefining the core agent runtime.
+
+## License
+
+Apache-2.0. See `LICENSE`.

{proseforge_agent-0.3.2 → proseforge_agent-0.4.0}/pyproject.toml

@@ -4,13 +4,18 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "proseforge-agent"
-version = "0.3.2"
-description = "Agentic orchestration layer for long-form novel production on top of the ProseForge engine."
+version = "0.4.0"
+description = "General agent runtime with a built-in ProseForge writing workflow pack."
 readme = "README.md"
-requires-python = ">=3.10"
+requires-python = ">=3.10,<3.13"
 license = { text = "Apache-2.0" }
 authors = [{ name = "ProseForge Agent" }]
-dependencies = ["pyyaml>=6.0", "tomli>=2.0; python_version < '3.11'"]
+dependencies = [
+    "pyyaml>=6.0",
+    "rich>=13.0",
+    "prompt_toolkit>=3.0",
+    "tomli>=2.0; python_version < '3.11'",
+]
 
 [project.optional-dependencies]
 dev = ["pytest>=7.0"]

{proseforge_agent-0.3.2 → proseforge_agent-0.4.0}/src/proseforge_agent/__init__.py

@@ -5,6 +5,6 @@ or require API keys. Subsystem behavior lives behind dedicated submodules
 that are imported explicitly by their callers.
 """
 
-__version__ = "0.3.2"
+__version__ = "0.4.0"
 
 __all__ = ["__version__"]

{proseforge_agent-0.3.2 → proseforge_agent-0.4.0}/src/proseforge_agent/agent/intent_router.py

@@ -8,6 +8,15 @@ from ..errors import ConfigurationError
 from .modes import CONVERSATION_MODES
 
 
+_TARGET_TOOLS = (
+    "install.doctor",
+    "provider.configure",
+    "workflow.continue",
+    "workflow.start",
+    "memory.add_candidate",
+)
+
+
 @dataclass(frozen=True)
 class IntentDecision:
     """Intent metadata returned to the Agent Kernel."""
@@ -149,4 +158,9 @@ class IntentRouter:
         return normalized in {"write it", "write"} or normalized.startswith("write ")
 
 
-__all__ = ["IntentDecision", "IntentRouter"]
+def intent_tool_targets() -> tuple[str, ...]:
+    """Return all tool ids the deterministic router can target."""
+    return _TARGET_TOOLS
+
+
+__all__ = ["IntentDecision", "IntentRouter", "intent_tool_targets"]

{proseforge_agent-0.3.2 → proseforge_agent-0.4.0}/src/proseforge_agent/agent/kernel.py

@@ -33,6 +33,18 @@ _PERMISSION_ORDER = {
 }
 
 
+def _render_evidence(evidence: list[dict[str, Any]]) -> str:
+    lines: list[str] = []
+    for item in evidence:
+        ref = str(item.get("id", "evidence"))
+        text = " ".join(str(item.get("text", "")).split())
+        if text:
+            lines.append(f"- [{ref}] {text}")
+        else:
+            lines.append(f"- [{ref}]")
+    return "\n".join(lines)
+
+
 class AgentKernel:
     """Run one agent turn through classification, retrieval, tools, and provider."""
 
@@ -89,7 +101,7 @@ class AgentKernel:
                 permission_level=effective_ceiling,
             )
 
-        text = self._call_provider(request, intent, events, trace_id, evidence_refs)
+        text = self._call_provider(request, intent, events, trace_id, evidence)
         self._append_message(request.session_id, "assistant", text)
         self._record_events(events)
         return AgentTurnResult(
@@ -142,8 +154,8 @@ class AgentKernel:
             return
 
         content = request.text
-        if evidence_refs:
-            content = f"{request.text}\nEvidence refs: {', '.join(evidence_refs)}"
+        if evidence:
+            content = f"{request.text}\n\nProject evidence:\n{_render_evidence(evidence)}"
         provider_request = ProviderRequest(
             role="planner" if intent.name == "retrieve_context" else "drafter",
             messages=self._provider_messages(request, content),
@@ -319,6 +331,32 @@ class AgentKernel:
             arguments = dict(rewritten.arguments)
 
         required = self._tool_permission(tool_name) or intent.required_permission
+        if self._tools is None:
+            events.append(
+                {
+                    "type": "tool_registry_missing",
+                    "trace_id": trace_id,
+                    "tool": tool_name,
+                }
+            )
+            text = f"Tool registry is not configured for this chat surface. Trace: {trace_id}"
+            self._append_message(request.session_id, "assistant", text)
+            self._record_events(events)
+            return AgentTurnResult(
+                text=text,
+                intent=intent,
+                tool_calls=[
+                    ToolCallResult(
+                        name=tool_name or "",
+                        status="failed",
+                        error="tool registry missing",
+                    )
+                ],
+                evidence_refs=evidence_refs,
+                memory_candidate_ids=memory_ids,
+                events=events,
+                trace_id=trace_id,
+            )
         if not self._permission_allows(permission_level, required):
             events.append(
                 {
@@ -326,9 +364,13 @@ class AgentKernel:
                     "trace_id": trace_id,
                     "tool": tool_name,
                     "required_permission": required,
+                    "current_permission": permission_level,
                 }
             )
-            text = f"Permission denied for {tool_name}; requires {required}. Trace {trace_id}."
+            text = (
+                f"Permission denied for {tool_name}; "
+                f"required={required} current={permission_level}. Trace {trace_id}."
+            )
             self._append_message(request.session_id, "assistant", text)
             self._record_events(events)
             return AgentTurnResult(
@@ -400,11 +442,11 @@ class AgentKernel:
         intent: AgentIntent,
         events: list[dict[str, Any]],
         trace_id: str,
-        evidence_refs: list[str],
+        evidence: list[dict[str, Any]],
     ) -> str:
         content = request.text
-        if evidence_refs:
-            content = f"{request.text}\nEvidence refs: {', '.join(evidence_refs)}"
+        if evidence:
+            content = f"{request.text}\n\nProject evidence:\n{_render_evidence(evidence)}"
         provider_request = ProviderRequest(
             role="planner" if intent.name == "retrieve_context" else "drafter",
             messages=self._provider_messages(request, content),

{proseforge_agent-0.3.2 → proseforge_agent-0.4.0}/src/proseforge_agent/agent/tools.py

@@ -359,6 +359,7 @@ def default_tool_registry() -> ToolRegistry:
         ("chapter.prepare", "draft_write", "Prepare chapter context"),
         ("chapter.run", "draft_write", "Draft chapter artifacts"),
         ("chapter.accept", "project_write", "Accept chapter output"),
+        ("provider.configure", "draft_write", "Configure a model provider"),
         ("provider.certify", "draft_write", "Write provider certification records"),
         ("install.doctor", "read_only", "Run installation diagnostics"),
         ("install.shell_completion", "system_write", "Install shell completions"),

{proseforge_agent-0.3.2 → proseforge_agent-0.4.0}/src/proseforge_agent/chat/__init__.py

@@ -4,6 +4,7 @@ from .handoff import ChatWorkflowHandoff, HandoffPackage
 from .context import ActiveContext, ActiveContextStore
 from .memory import ChatMemoryExtractor, MemoryCandidate
 from .prompts import ChatPromptBuilder, PromptItem, PromptPack
+from .project_retrieval import ProjectChatRetriever
 from .retrieval import ChatAnswer, ChatCitation, ChatEvidence, ChatRetrievalResponder
 from .session import (
     ChatContext,
@@ -41,6 +42,7 @@ __all__ = [
     "MemoryCandidate",
     "PromptItem",
     "PromptPack",
+    "ProjectChatRetriever",
     "SlashCommandAction",
     "SlashCommandContext",
     "SlashCommandRegistry",

proseforge_agent-0.4.0/src/proseforge_agent/chat/project_retrieval.py

@@ -0,0 +1,127 @@
+"""Lightweight project evidence retrieval for chat surfaces."""
+
+from __future__ import annotations
+
+import re
+from pathlib import Path
+from typing import Any
+
+
+_TEXT_SUFFIXES = {".md", ".txt", ".yaml", ".yml", ".json", ".jsonl"}
+_SOURCE_HINTS = (
+    ("manifest", ("manifest",)),
+    ("bible", ("bible",)),
+    ("timeline", ("timeline",)),
+    ("relation", ("relation", "relationship")),
+    ("foreshadow", ("foreshadow",)),
+    ("character_arc", ("character", "arc")),
+    ("chapter", ("chapter", "chapters", "scene", "scenes")),
+    ("rag", ("rag", "chunk", "chunks")),
+)
+
+
+class ProjectChatRetriever:
+    """Scan project artifacts and return deterministic evidence snippets.
+
+    This first pass intentionally avoids vector dependencies. It favours
+    predictable file-structure hints and keyword overlap so project chat has
+    useful citations even in a fresh PyPI install.
+    """
+
+    def __init__(self, workspace_root: str | Path = ".pf-agent/workspace", *, max_results: int = 8) -> None:
+        self.workspace_root = Path(workspace_root)
+        self.max_results = max(1, max_results)
+
+    def retrieve(self, project_slug: str | None, query: str) -> list[dict[str, Any]]:
+        if not project_slug:
+            return []
+        project_root = (self.workspace_root / project_slug).resolve()
+        try:
+            project_root.relative_to(self.workspace_root.resolve())
+        except ValueError:
+            return []
+        if not project_root.is_dir():
+            return []
+
+        tokens = _tokens(query)
+        candidates: list[dict[str, Any]] = []
+        for path in sorted(project_root.rglob("*")):
+            if not path.is_file() or path.suffix.lower() not in _TEXT_SUFFIXES:
+                continue
+            try:
+                text = path.read_text(encoding="utf-8")
+            except UnicodeDecodeError:
+                continue
+            rel_path = path.relative_to(project_root)
+            source = _source_for(rel_path)
+            score, snippet = _score_text(text, tokens, source)
+            if score <= 0:
+                continue
+            candidates.append(
+                {
+                    "id": f"{source}:{rel_path.as_posix()}",
+                    "source": source,
+                    "text": snippet,
+                    "path": str(path),
+                    "confidence": round(min(0.99, score / 12), 2),
+                    "score": score,
+                }
+            )
+        candidates.sort(key=lambda item: (-float(item["score"]), str(item["path"])))
+        return [
+            {key: value for key, value in item.items() if key != "score"}
+            for item in candidates[: self.max_results]
+        ]
+
+
+def _tokens(text: str) -> set[str]:
+    lowered = text.lower()
+    tokens = {token for token in re.findall(r"[a-z0-9_]{2,}", lowered)}
+    tokens.update(re.findall(r"[\u4e00-\u9fff]{2,}", text))
+    return tokens
+
+
+def _source_for(path: Path) -> str:
+    haystack = path.as_posix().lower()
+    for source, hints in _SOURCE_HINTS:
+        if any(hint in haystack for hint in hints):
+            return source
+    return "artifact"
+
+
+def _score_text(text: str, tokens: set[str], source: str) -> tuple[float, str]:
+    lines = [line.strip() for line in text.splitlines() if line.strip()]
+    if not lines:
+        return 0, ""
+    source_bonus = {
+        "manifest": 1.5,
+        "bible": 2.0,
+        "timeline": 2.0,
+        "relation": 1.8,
+        "foreshadow": 1.8,
+        "character_arc": 1.8,
+        "chapter": 1.4,
+        "rag": 1.6,
+    }.get(source, 1.0)
+    best_line = lines[0]
+    best_overlap = 0
+    for line in lines:
+        line_tokens = _tokens(line)
+        overlap = len(tokens & line_tokens) if tokens else 0
+        if overlap > best_overlap:
+            best_overlap = overlap
+            best_line = line
+    if best_overlap == 0:
+        # Keep high-signal project structure lightly discoverable.
+        if source in {"manifest", "bible", "timeline", "relation", "foreshadow", "character_arc"}:
+            return source_bonus, _snippet(best_line)
+        return 0, ""
+    return best_overlap + source_bonus, _snippet(best_line)
+
+
+def _snippet(text: str, *, limit: int = 360) -> str:
+    clean = " ".join(text.split())
+    return clean if len(clean) <= limit else clean[: limit - 1].rstrip() + "..."
+
+
+__all__ = ["ProjectChatRetriever"]

{proseforge_agent-0.3.2 → proseforge_agent-0.4.0}/src/proseforge_agent/chat/repl.py

@@ -7,7 +7,9 @@ from .. import _bootstrap  # noqa: F401  # UTF-8 + path hardening, must run earl
 import sys
 from typing import TextIO
 
-from ..agent import AgentKernel, AgentTurnRequest, IntentRouter
+from ..agent import AgentKernel, AgentTurnRequest, IntentRouter, default_tool_registry
+from ..agent.safety import InjectionGuard
+from .project_retrieval import ProjectChatRetriever
 from .session import ChatSessionStore
 from .slash import SlashCommandContext, SlashCommandRegistry
 
@@ -40,8 +42,11 @@ class ChatRepl:
         self.slash_registry = SlashCommandRegistry.default()
         self.kernel = AgentKernel(
             provider=provider,
+            tools=default_tool_registry(),
+            retrieval=ProjectChatRetriever(session_store.root / "workspace") if project_slug else None,
             session_store=session_store,
             intent_router=IntentRouter(),
+            safety=InjectionGuard(),
         )
 
     def run(self) -> int: