fableforge-anvil-agent 0.1.0__tar.gz

fableforge_anvil_agent-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2025 FableForge Contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

fableforge_anvil_agent-0.1.0/PKG-INFO

@@ -0,0 +1,289 @@
+Metadata-Version: 2.4
+Name: fableforge-anvil-agent
+Version: 0.1.0
+Summary: The open-source, self-verified coding agent. Generate → Execute → Verify → Recover.
+Author-email: FableForge <team@fableforge.ai>
+License: MIT
+Keywords: agent,coding,verification,self-healing,llm,open-source
+Classifier: Development Status :: 4 - Beta
+Classifier: Environment :: Console
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Software Development :: Code Generators
+Classifier: Topic :: Software Development :: Testing
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: rich>=13.0
+Requires-Dist: click>=8.1
+Requires-Dist: pydantic>=2.0
+Requires-Dist: httpx>=0.25
+Requires-Dist: tiktoken>=0.5
+Requires-Dist: pathspec>=0.11
+Requires-Dist: tree-sitter>=0.20
+Requires-Dist: prompt-toolkit>=3.0
+Requires-Dist: jsonschema>=4.0
+Requires-Dist: python-slugify>=8.0
+Provides-Extra: local
+Requires-Dist: llama-cpp-python>=0.2; extra == "local"
+Requires-Dist: onnxruntime>=1.16; extra == "local"
+Provides-Extra: api
+Requires-Dist: openai>=1.0; extra == "api"
+Requires-Dist: anthropic>=0.18; extra == "api"
+Provides-Extra: tui
+Requires-Dist: textual>=0.47; extra == "tui"
+Requires-Dist: textual-dev>=0.4; extra == "tui"
+Provides-Extra: all
+Requires-Dist: fableforge-anvil-agent[api,local,tui]; extra == "all"
+Provides-Extra: dev
+Requires-Dist: pytest>=7.0; extra == "dev"
+Requires-Dist: pytest-asyncio>=0.21; extra == "dev"
+Requires-Dist: pytest-cov>=4.0; extra == "dev"
+Requires-Dist: ruff>=0.1; extra == "dev"
+Dynamic: license-file
+
+<picture>
+  <source media="(prefers-color-scheme: dark)" srcset="docs/assets/logo-dark.svg">
+  <source media="(prefers-color-scheme: light)" srcset="docs/assets/logo-light.svg">
+  <img alt="Anvil" src="docs/assets/logo-light.svg" width="400">
+</picture>
+
+# Anvil — The Self-Verified Coding Agent
+
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE) [![Python 3.10+](https://img.shields.io/badge/python-3.10%2B-blue.svg)](https://www.python.org/downloads/) [![Tests](https://img.shields.io/badge/tests-278+-green.svg)](tests/)
+
+
+> **Generate → Execute → Verify → Recover**
+
+Every other open agent generates and hopes. **Anvil generates, runs, checks, and fixes** — because it was trained on 210,000 examples of real agents doing exactly that.
+
+This isn't prompt engineering. This is **behavior engineering**.
+
+---
+
+## Why Anvil?
+
+| Other Agents | Anvil |
+|---|---|
+| Generate code and hope it works | Generate code, then **verify it works** |
+| No error recovery | **Self-healing** with 3 retry attempts |
+| One-shot output | **Iterative** Plan→Execute→Verify→Recover loop |
+| No cost awareness | **Token tracking + model routing** for cost optimization |
+| Black box | **Full session tracking**, verify reports, telemetry |
+| Requires expensive API | Runs **fully local** with ShellWhisperer (1.5B) |
+
+## The Verification Loop
+
+```
+   ┌──────┐     ┌──────┐     ┌──────┐     ┌──────┐
+   │ PLAN │────▶│ EXEC │────▶│VERIFY│────▶│ DONE │
+   └──────┘     └──────┘     └──┬───┘     └──────┘
+                                 │ Fail
+                                 ▼
+                            ┌──────┐
+                            │RECOVR│────▶ back to EXEC
+                            └──────┘
+```
+
+Anvil doesn't just write code. It **verifies** every change:
+
+1. **Syntax check** — Does the code parse?
+2. **Test run** — Do the tests pass?
+3. **Lint check** — Is the code clean?
+4. **Import check** — Are dependencies valid?
+
+If verification fails, Anvil **diagnoses the error, generates a fix, and re-verifies**. Up to 3 retry cycles. This isn't optional — it's the core loop.
+
+## Quick Start
+
+```bash
+pip install anvil-agent
+
+# Run with local model (ollama)
+anvil run "Add error handling to main.py"
+
+# Run with API model
+anvil run -m gpt-4o "Refactor the auth module"
+
+# Interactive chat with verification
+anvil chat
+
+# Verify existing code
+anvil verify src/
+
+# Start as persistent daemon
+anvil daemon --port 8765
+
+# List past sessions
+anvil sessions
+```
+
+## The Name
+
+**Anvil** — where code gets forged, hammered, and tested until it holds.
+
+Every blacksmith knows: you don't just shape metal on the anvil. You **test** it. You strike it, check it, and if it's not right, you heat it again and hammer it until it is. That's what this agent does with code.
+
+**Other agents shape and ship. Anvil shapes, verifies, and only then ships.**
+
+## Architecture
+
+```
+anvil/
+├── core/
+│   ├── engine.py          # Plan→Execute→Verify→Recover loop
+│   ├── config.py          # 7-layer configuration system
+│   └── session.py          # Full session tracking + persistence
+├── tools/
+│   └── executor.py         # Bash, Read, Write, Edit, Grep, Glob, LS
+├── verify/
+│   └── pipeline.py         # Syntax, test, lint, import verification
+├── models/
+│   └── registry.py          # Local (ollama), OpenAI, Anthropic + cost tracking
+├── integrations/
+│   ├── verifyloop.py        # VerifyLoop framework integration
+│   ├── error_recovery.py   # ErrorRecovery engine integration
+│   ├── agent_swarm.py      # AgentSwarm coordination integration
+│   └── cost_optimizer.py   # CostOptimizer routing integration
+├── daemon/
+│   └── server.py            # Persistent HTTP daemon mode
+├── tui/
+│   └── dashboard.py         # Rich terminal dashboard
+└── cli.py                  # run, chat, verify, daemon, sessions, models
+```
+
+## The FableForge Ecosystem
+
+Anvil is the flagship product of the **FableForge** ecosystem — 21 open-source projects built from 210K real agent traces:
+
+| Project | What It Does |
+|---|---|
+| **Anvil** | Self-verified coding agent (this one) |
+| VerifyLoop | Plan→Execute→Verify→Recover framework |
+| ErrorRecovery | Self-healing middleware (3,725 error examples) |
+| FableForge-14B | The fine-tuned model (4-stage training) |
+| ShellWhisperer | 1.5B edge agent (phone/RPi, 50ms) |
+| ReasonCritic | Verification model (130 benchmark tasks) |
+| TraceCompiler | Compile traces → LoRA skills |
+| AgentRuntime | Persistent agent daemon (systemd for AI) |
+| AgentSwarm | Multi-agent from real trace transitions |
+| AgentTelemetry | Datadog for agents (token tracking, costs) |
+| BenchAgent | HumanEval for tool-use (107 tasks) |
+| AgentDev | VSCode extension with verification |
+| TraceViz | Trace replay visualizer (Next.js) |
+| AgentSkills.org | npm for agent behaviors |
+| AgentCurriculum | 5-stage progressive training |
+| AgentFuzzer | Adversarial testing for agents |
+| AgentConstitution | Safety guardrails from traces |
+| CostOptimizer | Token cost reduction (50-80%) |
+| AgentProfiler | Behavioral fingerprinting |
+| TrajectoryDistiller | Trace→training data pipeline |
+| Fable5-Dataset | HuggingFace dataset release |
+
+## Configuration
+
+Create `.anvil.json` in your project root:
+
+```json
+{
+  "model": {
+    "model": "local",
+    "temperature": 0.2,
+    "max_tokens": 4096
+  },
+  "verify": {
+    "enabled": true,
+    "auto_recover": true,
+    "max_retries": 3,
+    "check_syntax": true,
+    "check_tests": true,
+    "check_lint": true
+  },
+  "tools": {
+    "allow_shell": true,
+    "sandbox": false
+  },
+  "safety": {
+    "constitution_enabled": true,
+    "blocked_commands": ["rm -rf /", "mkfs"],
+    "require_confirmation_for": ["git push", "DROP TABLE"]
+  },
+  "cost": {
+    "max_cost_per_session_usd": 5.0,
+    "route_by_complexity": true,
+    "simple_model": "local",
+    "complex_model": "gpt-4o"
+  }
+}
+```
+
+## Daemon Mode
+
+Run Anvil as a persistent server:
+
+```bash
+anvil daemon --port 8765
+```
+
+```bash
+curl -X POST http://localhost:8765/run \
+  -H "Content-Type: application/json" \
+  -d '{"task": "Add input validation to all API endpoints"}'
+```
+
+## Model Backends
+
+| Model | Type | Input $/1M | Output $/1M |
+|---|---|---|---|
+| local (fableforge-14b) | Local | Free | Free |
+| gpt-4o | API | $2.50 | $10.00 |
+| gpt-4o-mini | API | $0.15 | $0.60 |
+| o3-mini | API | $1.10 | $4.40 |
+| claude-3.5-sonnet | API | $3.00 | $15.00 |
+| claude-3.5-haiku | API | $0.80 | $4.00 |
+
+## How It's Different
+
+### Trained on Real Behavior
+
+The FableForge model was trained on 210K examples from real agent traces:
+- **87.7% planning rate** — agents plan before they act
+- **39.5% error recovery rate** — agents that hit errors and recover
+- **1,311-step trace** — the Boeing 747 trace proves agents need persistent runtime
+- **31 tools** mapped — transition matrices drive swarm coordination
+
+### Verification Is Not Optional
+
+Other agents: "Here's the code, hope it works."
+
+Anvil: "Here's the code. I ran it. Tests pass. Lint is clean. Imports resolve. Here's the proof."
+
+### Self-Healing
+
+When verification fails, Anvil doesn't just report the error. It **reads the error, generates a fix, applies it, and re-verifies**. This is the ErrorRecovery engine with 3,725 real error examples baked in.
+
+### Ecosystem Integration
+
+Anvil doesn't work alone. It's wired into the full FableForge stack:
+- **VerifyLoop** → Sophisticated multi-step verification
+- **ErrorRecovery** → Pattern-matched error resolution from real traces
+- **AgentSwarm** → Multi-agent coordination via transition matrices
+- **CostOptimizer** → Automatic model routing based on task complexity
+- **AgentConstitution** → Safety guardrails from analysis of real traces
+
+## License
+
+MIT
+
+## Built With
+
+- 210,000+ real agent traces from the Fable-5 dataset collection
+- 87.7% planning rate behavioral signal
+- 39.5% error recovery success rate
+- 303 tool calls in a single session (Boeing 747 trace)
+- 5 specialized micro-models (ShellWhisperer, ReasonCritic, etc.)
+
+---
+
+**Anvil: Forge your code. Verify it holds.**

fableforge_anvil_agent-0.1.0/README.md

@@ -0,0 +1,244 @@
+<picture>
+  <source media="(prefers-color-scheme: dark)" srcset="docs/assets/logo-dark.svg">
+  <source media="(prefers-color-scheme: light)" srcset="docs/assets/logo-light.svg">
+  <img alt="Anvil" src="docs/assets/logo-light.svg" width="400">
+</picture>
+
+# Anvil — The Self-Verified Coding Agent
+
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE) [![Python 3.10+](https://img.shields.io/badge/python-3.10%2B-blue.svg)](https://www.python.org/downloads/) [![Tests](https://img.shields.io/badge/tests-278+-green.svg)](tests/)
+
+
+> **Generate → Execute → Verify → Recover**
+
+Every other open agent generates and hopes. **Anvil generates, runs, checks, and fixes** — because it was trained on 210,000 examples of real agents doing exactly that.
+
+This isn't prompt engineering. This is **behavior engineering**.
+
+---
+
+## Why Anvil?
+
+| Other Agents | Anvil |
+|---|---|
+| Generate code and hope it works | Generate code, then **verify it works** |
+| No error recovery | **Self-healing** with 3 retry attempts |
+| One-shot output | **Iterative** Plan→Execute→Verify→Recover loop |
+| No cost awareness | **Token tracking + model routing** for cost optimization |
+| Black box | **Full session tracking**, verify reports, telemetry |
+| Requires expensive API | Runs **fully local** with ShellWhisperer (1.5B) |
+
+## The Verification Loop
+
+```
+   ┌──────┐     ┌──────┐     ┌──────┐     ┌──────┐
+   │ PLAN │────▶│ EXEC │────▶│VERIFY│────▶│ DONE │
+   └──────┘     └──────┘     └──┬───┘     └──────┘
+                                 │ Fail
+                                 ▼
+                            ┌──────┐
+                            │RECOVR│────▶ back to EXEC
+                            └──────┘
+```
+
+Anvil doesn't just write code. It **verifies** every change:
+
+1. **Syntax check** — Does the code parse?
+2. **Test run** — Do the tests pass?
+3. **Lint check** — Is the code clean?
+4. **Import check** — Are dependencies valid?
+
+If verification fails, Anvil **diagnoses the error, generates a fix, and re-verifies**. Up to 3 retry cycles. This isn't optional — it's the core loop.
+
+## Quick Start
+
+```bash
+pip install anvil-agent
+
+# Run with local model (ollama)
+anvil run "Add error handling to main.py"
+
+# Run with API model
+anvil run -m gpt-4o "Refactor the auth module"
+
+# Interactive chat with verification
+anvil chat
+
+# Verify existing code
+anvil verify src/
+
+# Start as persistent daemon
+anvil daemon --port 8765
+
+# List past sessions
+anvil sessions
+```
+
+## The Name
+
+**Anvil** — where code gets forged, hammered, and tested until it holds.
+
+Every blacksmith knows: you don't just shape metal on the anvil. You **test** it. You strike it, check it, and if it's not right, you heat it again and hammer it until it is. That's what this agent does with code.
+
+**Other agents shape and ship. Anvil shapes, verifies, and only then ships.**
+
+## Architecture
+
+```
+anvil/
+├── core/
+│   ├── engine.py          # Plan→Execute→Verify→Recover loop
+│   ├── config.py          # 7-layer configuration system
+│   └── session.py          # Full session tracking + persistence
+├── tools/
+│   └── executor.py         # Bash, Read, Write, Edit, Grep, Glob, LS
+├── verify/
+│   └── pipeline.py         # Syntax, test, lint, import verification
+├── models/
+│   └── registry.py          # Local (ollama), OpenAI, Anthropic + cost tracking
+├── integrations/
+│   ├── verifyloop.py        # VerifyLoop framework integration
+│   ├── error_recovery.py   # ErrorRecovery engine integration
+│   ├── agent_swarm.py      # AgentSwarm coordination integration
+│   └── cost_optimizer.py   # CostOptimizer routing integration
+├── daemon/
+│   └── server.py            # Persistent HTTP daemon mode
+├── tui/
+│   └── dashboard.py         # Rich terminal dashboard
+└── cli.py                  # run, chat, verify, daemon, sessions, models
+```
+
+## The FableForge Ecosystem
+
+Anvil is the flagship product of the **FableForge** ecosystem — 21 open-source projects built from 210K real agent traces:
+
+| Project | What It Does |
+|---|---|
+| **Anvil** | Self-verified coding agent (this one) |
+| VerifyLoop | Plan→Execute→Verify→Recover framework |
+| ErrorRecovery | Self-healing middleware (3,725 error examples) |
+| FableForge-14B | The fine-tuned model (4-stage training) |
+| ShellWhisperer | 1.5B edge agent (phone/RPi, 50ms) |
+| ReasonCritic | Verification model (130 benchmark tasks) |
+| TraceCompiler | Compile traces → LoRA skills |
+| AgentRuntime | Persistent agent daemon (systemd for AI) |
+| AgentSwarm | Multi-agent from real trace transitions |
+| AgentTelemetry | Datadog for agents (token tracking, costs) |
+| BenchAgent | HumanEval for tool-use (107 tasks) |
+| AgentDev | VSCode extension with verification |
+| TraceViz | Trace replay visualizer (Next.js) |
+| AgentSkills.org | npm for agent behaviors |
+| AgentCurriculum | 5-stage progressive training |
+| AgentFuzzer | Adversarial testing for agents |
+| AgentConstitution | Safety guardrails from traces |
+| CostOptimizer | Token cost reduction (50-80%) |
+| AgentProfiler | Behavioral fingerprinting |
+| TrajectoryDistiller | Trace→training data pipeline |
+| Fable5-Dataset | HuggingFace dataset release |
+
+## Configuration
+
+Create `.anvil.json` in your project root:
+
+```json
+{
+  "model": {
+    "model": "local",
+    "temperature": 0.2,
+    "max_tokens": 4096
+  },
+  "verify": {
+    "enabled": true,
+    "auto_recover": true,
+    "max_retries": 3,
+    "check_syntax": true,
+    "check_tests": true,
+    "check_lint": true
+  },
+  "tools": {
+    "allow_shell": true,
+    "sandbox": false
+  },
+  "safety": {
+    "constitution_enabled": true,
+    "blocked_commands": ["rm -rf /", "mkfs"],
+    "require_confirmation_for": ["git push", "DROP TABLE"]
+  },
+  "cost": {
+    "max_cost_per_session_usd": 5.0,
+    "route_by_complexity": true,
+    "simple_model": "local",
+    "complex_model": "gpt-4o"
+  }
+}
+```
+
+## Daemon Mode
+
+Run Anvil as a persistent server:
+
+```bash
+anvil daemon --port 8765
+```
+
+```bash
+curl -X POST http://localhost:8765/run \
+  -H "Content-Type: application/json" \
+  -d '{"task": "Add input validation to all API endpoints"}'
+```
+
+## Model Backends
+
+| Model | Type | Input $/1M | Output $/1M |
+|---|---|---|---|
+| local (fableforge-14b) | Local | Free | Free |
+| gpt-4o | API | $2.50 | $10.00 |
+| gpt-4o-mini | API | $0.15 | $0.60 |
+| o3-mini | API | $1.10 | $4.40 |
+| claude-3.5-sonnet | API | $3.00 | $15.00 |
+| claude-3.5-haiku | API | $0.80 | $4.00 |
+
+## How It's Different
+
+### Trained on Real Behavior
+
+The FableForge model was trained on 210K examples from real agent traces:
+- **87.7% planning rate** — agents plan before they act
+- **39.5% error recovery rate** — agents that hit errors and recover
+- **1,311-step trace** — the Boeing 747 trace proves agents need persistent runtime
+- **31 tools** mapped — transition matrices drive swarm coordination
+
+### Verification Is Not Optional
+
+Other agents: "Here's the code, hope it works."
+
+Anvil: "Here's the code. I ran it. Tests pass. Lint is clean. Imports resolve. Here's the proof."
+
+### Self-Healing
+
+When verification fails, Anvil doesn't just report the error. It **reads the error, generates a fix, applies it, and re-verifies**. This is the ErrorRecovery engine with 3,725 real error examples baked in.
+
+### Ecosystem Integration
+
+Anvil doesn't work alone. It's wired into the full FableForge stack:
+- **VerifyLoop** → Sophisticated multi-step verification
+- **ErrorRecovery** → Pattern-matched error resolution from real traces
+- **AgentSwarm** → Multi-agent coordination via transition matrices
+- **CostOptimizer** → Automatic model routing based on task complexity
+- **AgentConstitution** → Safety guardrails from analysis of real traces
+
+## License
+
+MIT
+
+## Built With
+
+- 210,000+ real agent traces from the Fable-5 dataset collection
+- 87.7% planning rate behavioral signal
+- 39.5% error recovery success rate
+- 303 tool calls in a single session (Boeing 747 trace)
+- 5 specialized micro-models (ShellWhisperer, ReasonCritic, etc.)
+
+---
+
+**Anvil: Forge your code. Verify it holds.**

fableforge_anvil_agent-0.1.0/pyproject.toml

@@ -0,0 +1,64 @@
+[build-system]
+requires = ["setuptools>=68.0", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "fableforge-anvil-agent"
+version = "0.1.0"
+description = "The open-source, self-verified coding agent. Generate → Execute → Verify → Recover."
+readme = "README.md"
+license = {text = "MIT"}
+requires-python = ">=3.10"
+authors = [{name = "FableForge", email = "team@fableforge.ai"}]
+keywords = ["agent", "coding", "verification", "self-healing", "llm", "open-source"]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Environment :: Console",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: MIT License",
+    "Programming Language :: Python :: 3",
+    "Topic :: Software Development :: Code Generators",
+    "Topic :: Software Development :: Testing",
+]
+
+dependencies = [
+    "rich>=13.0",
+    "click>=8.1",
+    "pydantic>=2.0",
+    "httpx>=0.25",
+    "tiktoken>=0.5",
+    "pathspec>=0.11",
+    "tree-sitter>=0.20",
+    "prompt-toolkit>=3.0",
+    "jsonschema>=4.0",
+    "python-slugify>=8.0",
+]
+
+[project.optional-dependencies]
+local = ["llama-cpp-python>=0.2", "onnxruntime>=1.16"]
+api = ["openai>=1.0", "anthropic>=0.18"]
+tui = ["textual>=0.47", "textual-dev>=0.4"]
+all = ["fableforge-anvil-agent[local,api,tui]"]
+dev = ["pytest>=7.0", "pytest-asyncio>=0.21", "pytest-cov>=4.0", "ruff>=0.1"]
+
+[project.scripts]
+anvil = "anvil.cli:main"
+
+[project.entry-points."anvil.models"]
+local = "anvil.models.local:LocalModel"
+openai = "anvil.models.openai_model:OpenAIModel"
+anthropic = "anvil.models.anthropic_model:AnthropicModel"
+
+[tool.setuptools.packages.find]
+where = ["src"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+asyncio_mode = "auto"
+
+[tool.ruff]
+target-version = "py310"
+line-length = 100
+
+[tool.ruff.lint]
+select = ["E", "F", "I", "N", "W", "UP"]

fableforge_anvil_agent-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+