cauterule 0.1.0__tar.gz

cauterule-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Debashish Ghosal
+
+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.

cauterule-0.1.0/PKG-INFO

@@ -0,0 +1,272 @@
+Metadata-Version: 2.4
+Name: cauterule
+Version: 0.1.0
+Summary: Automated standing-rule extraction from agent failures — extract, test, promote.
+Author: Debashish Ghosal
+Maintainer: Debashish Ghosal
+License: MIT
+Project-URL: Homepage, https://github.com/deghosal-2026/CauterRule
+Project-URL: Repository, https://github.com/deghosal-2026/CauterRule.git
+Project-URL: Issues, https://github.com/deghosal-2026/CauterRule/issues
+Keywords: agent,llm,rules,learning,memory,automation
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Typing :: Typed
+Requires-Python: >=3.11
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: click>=8.1
+Requires-Dist: pyyaml>=6.0
+Requires-Dist: textual>=1.0
+Requires-Dist: mcp<2,>=1.0
+Provides-Extra: dev
+Requires-Dist: pytest>=8.0; extra == "dev"
+Requires-Dist: pytest-cov>=5.0; extra == "dev"
+Requires-Dist: ruff>=0.8; extra == "dev"
+Requires-Dist: mypy>=1.13; extra == "dev"
+Requires-Dist: types-PyYAML>=6.0; extra == "dev"
+Requires-Dist: build>=1.0; extra == "dev"
+Requires-Dist: twine>=4.0; extra == "dev"
+Provides-Extra: llm
+Requires-Dist: openai>=1.0; extra == "llm"
+Requires-Dist: anthropic>=0.25; extra == "llm"
+Provides-Extra: all
+Requires-Dist: openai>=1.0; extra == "all"
+Requires-Dist: anthropic>=0.25; extra == "all"
+Dynamic: license-file
+
+# CauterRule
+
+**Automated standing-rule extraction from agent failures.**
+
+Every agent fails. CauterRule ensures they never make the *same* failure twice.
+
+After every failure, CauterRule:
+1. **Extracts** a structured *when X, do Y* standing rule from the execution trajectory (multi-pass with draft tournament)
+2. **Tests** that candidate rule against historical scenarios (past failures + past successes — deterministic replay)
+3. **Promotes** only rules that survive replay into the agent's permanent rule store (with linter, conflict detection, provenance)
+
+No more corrections dying in chat. No more hand-written standing rules. No more vague reflection paragraphs nobody re-reads. Rules are actionable, tested, and permanent.
+
+**Status:** v0.1.0 — Field test complete. [Read the full report](docs/field-test/v0.1.0/FIELD_TEST_REPORT.md).
+
+---
+
+## Why
+
+The current state of "learning from mistakes" in agent systems is reflection-as-a-paragraph — prose that bloats the context window and is never read again. The manual alternative is standing rules maintained by hand. Both are broken.
+
+CauterRule automates the extract → test → promote loop. Same pattern as CI/CD for code, but for agent behavior: test a rule against the historical regression suite before merging it in.
+
+---
+
+## What Makes It Different
+
+| Existing | CauterRule |
+|----------|-----------|
+| Reflection produces prose nobody re-reads | Produces structured *when X, do Y* rules that are testable and injectable |
+| Memory stores hold raw text (Letta/MemGPT) | Rules are tested against history before promotion — evidence-based, not append-only |
+| Framework memory is unstructured (LangGraph) | Rules have provenance, versioning, conflict detection, linter, and retirement |
+| Standing rules maintained by hand | The agent writes its own rules, tests them, and promotes only what survives |
+| `.cursorrules` / `CLAUDE.md` are static files | Rules are living artifacts that grow from real failures, not guesses |
+| No OSS tool ships a corpus or benchmarks | 13 corpus types, 394 trajectories, golden set, safety corpora, field test runner |
+
+---
+
+## Quick Start
+
+```bash
+pip install cauterule
+
+# Run the demo — seeded failures, full loop in 60s
+cauterule demo
+
+# Extract a rule from a trajectory
+cauterule extract trajectory.jsonl
+
+# Replay-test a candidate rule
+cauterule test R-001
+
+# Promote to the permanent store
+cauterule promote R-001
+
+# Browse rules
+cauterule list
+cauterule health
+cauterule validate
+
+# Export rules to your agent's format
+cauterule export --format agents
+```
+
+---
+
+## What's Shipped in v0.1.0
+
+### Core Loop
+- Trajectory capture with secret redaction
+- Multi-pass LLM extraction (3x, temperature variation, draft tournament)
+- Failure clustering — one extraction per failure cluster, not per failure
+- Historical replay engine with deterministic evidence reports
+- Promotion gate — auto, human-review, or hybrid mode
+- Rule linter (vagueness, tautology, duplicate, contradiction, untestable, unsafe)
+- Conflict detection and rule consolidation
+- Versioned YAML rule store with git provenance
+
+### CLI (25+ Commands)
+- `cauterule init` | `demo` | `extract` (`--dry-run`) | `test` (`--ci`) | `promote` | `inject` | `list` | `show` | `search`
+- `cauterule audit` | `diff` | `retire` | `history` | `conflicts` | `health` | `validate`
+- `cauterule counterfactual` | `story` | `explain` | `config` | `metrics` | `report` | `pack list` | `pack info`
+- `cauterule rewind` (Failure Time Machine) | `cauterule mcp` (MCP server)
+
+### Replay Engine
+- Replay harness with evidence reports (failures prevented, successes broken, precision, recall, verdict)
+- "What if?" mode — apply a hypothetical rule and simulate the outcome
+- Failure Time Machine — `cauterule rewind <trajectory>` with rule overlay
+- Rule Draft Tournament — generate candidates, replay all, rank, promote the winner
+
+### Rule Store
+- YAML rule files with provenance metadata and tags
+- Auto-classified failure taxonomy (`git/push`, `python/import`, `docker/network`)
+- Git-based versioning, rollback, archive directory
+- Rule consolidation — merge overlapping triggers
+
+### Rule Injection
+- Structured matching by trigger, tool, error type, context, tags, taxonomy
+- Specificity ordering — more specific rules injected first
+- Rule explanations, templating (retry, verify-then-act, check-preconditions)
+- Context budget optimizer, preflight mode
+
+### Rule Packs (Bundled)
+- `pack-git` — 10+ pre-built git rules (push, merge, rebase, conflicts)
+- Zero cold-start: rules work out of the box, zero LLM cost
+
+### Export & Import (Day-One Interop)
+- Export to `.cursorrules`, `CLAUDE.md`, `AGENTS.md`, `.windsurfrules`, `aider.conf.yml`, markdown, JSON
+- Import from `.cursorrules`, `CLAUDE.md`, `AGENTS.md`, and chat history
+
+### MCP Server
+- 4 tools: `get_matching_rules`, `get_rule`, `list_rules`, `report_failure`
+- Any MCP-compatible agent (Claude, etc.) consumes rules with zero code changes
+
+### Custom Agent Adapter
+- `@cauterule.watch` decorator — wrap any agent function, auto-captures trajectories on failure
+- `cauterule.inject()` context manager — prep context with matching rules before task execution
+- Works with any Python agent — no framework lock-in
+
+### Corpus & Field Test Infrastructure
+- 13 corpus types totaling 394 trajectories (curated + raw)
+- Golden trajectory set with known expected rules
+- Safety corpora: `successes`, `failures/negative`, `nearmiss`, `noisy`, `corrections`
+- Raw corpora: OpenCode sessions, synthetic scenarios, CI failure logs, sibling-repo runs
+- Field test runner supporting local OMLX and cloud OpenRouter LLMs
+- 844+ deterministic tests passing, 104 Docker tests passing
+
+### Configuration
+- `cauterule.toml` — LLM provider, model, thresholds, mode, paths, redaction patterns
+- Environment variable support — `CAUTERULE_LLM_PROVIDER`, `CAUTERULE_MODEL`, etc.
+- LLM provider abstraction — OpenAI, Anthropic, Ollama, LiteLLM
+- Promotion mode — auto, human-review, hybrid
+
+---
+
+## Field Test Results
+
+The full v0.1.0 field test evaluated 4 models across 13 corpus types (394 trajectories):
+
+| Model | Type | Candidates | Pass | Inconclusive | Fail |
+|---|---|---:|---:|---:|---:|
+| Llama-3.2-3B-Instruct-4bit | Local OMLX | 379 | 72 | 189 | 118 |
+| Qwen3-4B-Instruct-2507-4bit | Local OMLX | 373 | 93 | 209 | 71 |
+| openai/gpt-4o-mini | Cloud OpenRouter | 394 | 77 | 248 | 69 |
+| meta-llama/llama-3.1-8b-instruct | Cloud OpenRouter | 392 | 123 | 168 | 101 |
+
+Key findings:
+- Parser and prompt fixes improved local-model parse reliability from ~30% to near 100%
+- `meta-llama/llama-3.1-8b-instruct` was the strongest cost-effective model tested
+- Safety corpora remain the hardest unsolved area across all models
+
+Full report: [`docs/field-test/v0.1.0/FIELD_TEST_REPORT.md`](docs/field-test/v0.1.0/FIELD_TEST_REPORT.md)
+
+---
+
+## Architecture
+
+```
+                         ┌──────────────────────────────────────────┐
+                         │              CLI / MCP                    │
+                         │  cauterule demo | extract | mcp          │
+                         └──────┬───────────────────────────┬───────┘
+                                │                           │
+                                ▼                           ▼
+┌─────────────────┐     ┌──────────────────┐     ┌───────────────────┐
+│ Trajectory      │ ──> │ Rule Extractor   │ ──> │ Historical Replay │
+│ Capture + Redact│     │ (multi-pass LLM) │     │ Engine + Viz      │
+└─────────────────┘     └──────────────────┘     └───────────────────┘
+       │                        │                         │
+       ▼                        ▼                         ▼
+┌─────────────────┐     ┌──────────────┐          ┌───────────────────┐
+│ Rule Injection  │ <── │ Standing-Rules│ <── ─── │ Promotion Gate    │
+│ + Explanations  │     │ Store (YAML)  │     │   │ + Linter + Conflicts│
+└─────────────────┘     └──────┬───────┘     │
+                               │             │
+                               ▼             ▼
+┌─────────────────┐     ┌──────────────────┐     ┌───────────────────┐
+│ Export / Import │     │ Corpus & Field   │     │ Basic Observability│
+│ 7 formats       │     │ Test Runner      │     │ Metrics + Report  │
+│ MCP server      │     │ 394 trajectories │     │ Health + Validate │
+└─────────────────┘     └──────────────────┘     └───────────────────┘
+```
+
+---
+
+## Roadmap
+
+| Version | Theme | Key Deliverable |
+|---------|-------|-----------------|
+| **v0.1.0** ✅ | Core Loop + DX | Full loop, 25+ CLI commands, MCP, export/import, packs, corpus, field tests |
+| **v0.2.0** | Distribution + Polish | TUI review, observability, adversarial corpora, Homebrew, GitHub Action, webhook, OTEL |
+| **v0.3.0** | Rule Pack Ecosystem | pack install/create/publish, official packs (docker, deploy, testing, python) |
+| **v0.4.0** | Deep Integrations | AgentObservatory, AgentEvalForge, LangSmith/Phoenix |
+| **v0.5.0** | Observability & Analytics | Web dashboard, trend lines, weekly digest |
+| **v0.6.0** | Advanced Retrieval | Semantic matching, hybrid matching, rule embedding index |
+| **v0.7.0** | Multi-Agent | Cross-agent transfer, shared registry, rule governance |
+
+---
+
+## Related Ecosystem
+
+CauterRule is the **learning layer** in an open-source agent infrastructure stack:
+
+| Project | Layer | How it relates |
+|---------|-------|----------------|
+| [planner-critic-engine](https://github.com/deghosal-2026/planner-critic-engine) | Prevention | Catches flawed plans *before* execution |
+| [agent-self-edit](https://github.com/deghosal-2026/agent-self-edit) | Prompt Learning | Twin learning engine — extracts *prompt edits* from failures |
+| [agent-tooltrust](https://github.com/deghosal-2026/agent-tooltrust) | Tool Safety | Runtime permission engine |
+| [agent-eval-forge](https://github.com/deghosal-2026/agent-eval-forge) | Evaluation | "pytest for agents" |
+| [ai-loopguard](https://github.com/deghosal-2026/ai-loopguard) | Circuit Breaker | Real-time failure handling |
+
+---
+
+## Documentation
+
+- [User Guide](docs/USER_GUIDE.md)
+- [Field Test Report](docs/field-test/v0.1.0/FIELD_TEST_REPORT.md)
+- [Release Notes](docs/release/v0.1.0/release-notes.md)
+- [Changelog](CHANGELOG.md)
+- [Docs Index](docs/README.md)
+- [PRD: Why](docs/design/prd/01-why.md)
+- [Architecture](docs/design/prd/02-architecture.md)
+- [WBS v0.1.0](docs/wbs/v0.1.0/wbs-v0.1.0-index.md)
+
+---
+
+## License
+
+MIT — see [LICENSE](LICENSE).

cauterule-0.1.0/README.md

@@ -0,0 +1,229 @@
+# CauterRule
+
+**Automated standing-rule extraction from agent failures.**
+
+Every agent fails. CauterRule ensures they never make the *same* failure twice.
+
+After every failure, CauterRule:
+1. **Extracts** a structured *when X, do Y* standing rule from the execution trajectory (multi-pass with draft tournament)
+2. **Tests** that candidate rule against historical scenarios (past failures + past successes — deterministic replay)
+3. **Promotes** only rules that survive replay into the agent's permanent rule store (with linter, conflict detection, provenance)
+
+No more corrections dying in chat. No more hand-written standing rules. No more vague reflection paragraphs nobody re-reads. Rules are actionable, tested, and permanent.
+
+**Status:** v0.1.0 — Field test complete. [Read the full report](docs/field-test/v0.1.0/FIELD_TEST_REPORT.md).
+
+---
+
+## Why
+
+The current state of "learning from mistakes" in agent systems is reflection-as-a-paragraph — prose that bloats the context window and is never read again. The manual alternative is standing rules maintained by hand. Both are broken.
+
+CauterRule automates the extract → test → promote loop. Same pattern as CI/CD for code, but for agent behavior: test a rule against the historical regression suite before merging it in.
+
+---
+
+## What Makes It Different
+
+| Existing | CauterRule |
+|----------|-----------|
+| Reflection produces prose nobody re-reads | Produces structured *when X, do Y* rules that are testable and injectable |
+| Memory stores hold raw text (Letta/MemGPT) | Rules are tested against history before promotion — evidence-based, not append-only |
+| Framework memory is unstructured (LangGraph) | Rules have provenance, versioning, conflict detection, linter, and retirement |
+| Standing rules maintained by hand | The agent writes its own rules, tests them, and promotes only what survives |
+| `.cursorrules` / `CLAUDE.md` are static files | Rules are living artifacts that grow from real failures, not guesses |
+| No OSS tool ships a corpus or benchmarks | 13 corpus types, 394 trajectories, golden set, safety corpora, field test runner |
+
+---
+
+## Quick Start
+
+```bash
+pip install cauterule
+
+# Run the demo — seeded failures, full loop in 60s
+cauterule demo
+
+# Extract a rule from a trajectory
+cauterule extract trajectory.jsonl
+
+# Replay-test a candidate rule
+cauterule test R-001
+
+# Promote to the permanent store
+cauterule promote R-001
+
+# Browse rules
+cauterule list
+cauterule health
+cauterule validate
+
+# Export rules to your agent's format
+cauterule export --format agents
+```
+
+---
+
+## What's Shipped in v0.1.0
+
+### Core Loop
+- Trajectory capture with secret redaction
+- Multi-pass LLM extraction (3x, temperature variation, draft tournament)
+- Failure clustering — one extraction per failure cluster, not per failure
+- Historical replay engine with deterministic evidence reports
+- Promotion gate — auto, human-review, or hybrid mode
+- Rule linter (vagueness, tautology, duplicate, contradiction, untestable, unsafe)
+- Conflict detection and rule consolidation
+- Versioned YAML rule store with git provenance
+
+### CLI (25+ Commands)
+- `cauterule init` | `demo` | `extract` (`--dry-run`) | `test` (`--ci`) | `promote` | `inject` | `list` | `show` | `search`
+- `cauterule audit` | `diff` | `retire` | `history` | `conflicts` | `health` | `validate`
+- `cauterule counterfactual` | `story` | `explain` | `config` | `metrics` | `report` | `pack list` | `pack info`
+- `cauterule rewind` (Failure Time Machine) | `cauterule mcp` (MCP server)
+
+### Replay Engine
+- Replay harness with evidence reports (failures prevented, successes broken, precision, recall, verdict)
+- "What if?" mode — apply a hypothetical rule and simulate the outcome
+- Failure Time Machine — `cauterule rewind <trajectory>` with rule overlay
+- Rule Draft Tournament — generate candidates, replay all, rank, promote the winner
+
+### Rule Store
+- YAML rule files with provenance metadata and tags
+- Auto-classified failure taxonomy (`git/push`, `python/import`, `docker/network`)
+- Git-based versioning, rollback, archive directory
+- Rule consolidation — merge overlapping triggers
+
+### Rule Injection
+- Structured matching by trigger, tool, error type, context, tags, taxonomy
+- Specificity ordering — more specific rules injected first
+- Rule explanations, templating (retry, verify-then-act, check-preconditions)
+- Context budget optimizer, preflight mode
+
+### Rule Packs (Bundled)
+- `pack-git` — 10+ pre-built git rules (push, merge, rebase, conflicts)
+- Zero cold-start: rules work out of the box, zero LLM cost
+
+### Export & Import (Day-One Interop)
+- Export to `.cursorrules`, `CLAUDE.md`, `AGENTS.md`, `.windsurfrules`, `aider.conf.yml`, markdown, JSON
+- Import from `.cursorrules`, `CLAUDE.md`, `AGENTS.md`, and chat history
+
+### MCP Server
+- 4 tools: `get_matching_rules`, `get_rule`, `list_rules`, `report_failure`
+- Any MCP-compatible agent (Claude, etc.) consumes rules with zero code changes
+
+### Custom Agent Adapter
+- `@cauterule.watch` decorator — wrap any agent function, auto-captures trajectories on failure
+- `cauterule.inject()` context manager — prep context with matching rules before task execution
+- Works with any Python agent — no framework lock-in
+
+### Corpus & Field Test Infrastructure
+- 13 corpus types totaling 394 trajectories (curated + raw)
+- Golden trajectory set with known expected rules
+- Safety corpora: `successes`, `failures/negative`, `nearmiss`, `noisy`, `corrections`
+- Raw corpora: OpenCode sessions, synthetic scenarios, CI failure logs, sibling-repo runs
+- Field test runner supporting local OMLX and cloud OpenRouter LLMs
+- 844+ deterministic tests passing, 104 Docker tests passing
+
+### Configuration
+- `cauterule.toml` — LLM provider, model, thresholds, mode, paths, redaction patterns
+- Environment variable support — `CAUTERULE_LLM_PROVIDER`, `CAUTERULE_MODEL`, etc.
+- LLM provider abstraction — OpenAI, Anthropic, Ollama, LiteLLM
+- Promotion mode — auto, human-review, hybrid
+
+---
+
+## Field Test Results
+
+The full v0.1.0 field test evaluated 4 models across 13 corpus types (394 trajectories):
+
+| Model | Type | Candidates | Pass | Inconclusive | Fail |
+|---|---|---:|---:|---:|---:|
+| Llama-3.2-3B-Instruct-4bit | Local OMLX | 379 | 72 | 189 | 118 |
+| Qwen3-4B-Instruct-2507-4bit | Local OMLX | 373 | 93 | 209 | 71 |
+| openai/gpt-4o-mini | Cloud OpenRouter | 394 | 77 | 248 | 69 |
+| meta-llama/llama-3.1-8b-instruct | Cloud OpenRouter | 392 | 123 | 168 | 101 |
+
+Key findings:
+- Parser and prompt fixes improved local-model parse reliability from ~30% to near 100%
+- `meta-llama/llama-3.1-8b-instruct` was the strongest cost-effective model tested
+- Safety corpora remain the hardest unsolved area across all models
+
+Full report: [`docs/field-test/v0.1.0/FIELD_TEST_REPORT.md`](docs/field-test/v0.1.0/FIELD_TEST_REPORT.md)
+
+---
+
+## Architecture
+
+```
+                         ┌──────────────────────────────────────────┐
+                         │              CLI / MCP                    │
+                         │  cauterule demo | extract | mcp          │
+                         └──────┬───────────────────────────┬───────┘
+                                │                           │
+                                ▼                           ▼
+┌─────────────────┐     ┌──────────────────┐     ┌───────────────────┐
+│ Trajectory      │ ──> │ Rule Extractor   │ ──> │ Historical Replay │
+│ Capture + Redact│     │ (multi-pass LLM) │     │ Engine + Viz      │
+└─────────────────┘     └──────────────────┘     └───────────────────┘
+       │                        │                         │
+       ▼                        ▼                         ▼
+┌─────────────────┐     ┌──────────────┐          ┌───────────────────┐
+│ Rule Injection  │ <── │ Standing-Rules│ <── ─── │ Promotion Gate    │
+│ + Explanations  │     │ Store (YAML)  │     │   │ + Linter + Conflicts│
+└─────────────────┘     └──────┬───────┘     │
+                               │             │
+                               ▼             ▼
+┌─────────────────┐     ┌──────────────────┐     ┌───────────────────┐
+│ Export / Import │     │ Corpus & Field   │     │ Basic Observability│
+│ 7 formats       │     │ Test Runner      │     │ Metrics + Report  │
+│ MCP server      │     │ 394 trajectories │     │ Health + Validate │
+└─────────────────┘     └──────────────────┘     └───────────────────┘
+```
+
+---
+
+## Roadmap
+
+| Version | Theme | Key Deliverable |
+|---------|-------|-----------------|
+| **v0.1.0** ✅ | Core Loop + DX | Full loop, 25+ CLI commands, MCP, export/import, packs, corpus, field tests |
+| **v0.2.0** | Distribution + Polish | TUI review, observability, adversarial corpora, Homebrew, GitHub Action, webhook, OTEL |
+| **v0.3.0** | Rule Pack Ecosystem | pack install/create/publish, official packs (docker, deploy, testing, python) |
+| **v0.4.0** | Deep Integrations | AgentObservatory, AgentEvalForge, LangSmith/Phoenix |
+| **v0.5.0** | Observability & Analytics | Web dashboard, trend lines, weekly digest |
+| **v0.6.0** | Advanced Retrieval | Semantic matching, hybrid matching, rule embedding index |
+| **v0.7.0** | Multi-Agent | Cross-agent transfer, shared registry, rule governance |
+
+---
+
+## Related Ecosystem
+
+CauterRule is the **learning layer** in an open-source agent infrastructure stack:
+
+| Project | Layer | How it relates |
+|---------|-------|----------------|
+| [planner-critic-engine](https://github.com/deghosal-2026/planner-critic-engine) | Prevention | Catches flawed plans *before* execution |
+| [agent-self-edit](https://github.com/deghosal-2026/agent-self-edit) | Prompt Learning | Twin learning engine — extracts *prompt edits* from failures |
+| [agent-tooltrust](https://github.com/deghosal-2026/agent-tooltrust) | Tool Safety | Runtime permission engine |
+| [agent-eval-forge](https://github.com/deghosal-2026/agent-eval-forge) | Evaluation | "pytest for agents" |
+| [ai-loopguard](https://github.com/deghosal-2026/ai-loopguard) | Circuit Breaker | Real-time failure handling |
+
+---
+
+## Documentation
+
+- [User Guide](docs/USER_GUIDE.md)
+- [Field Test Report](docs/field-test/v0.1.0/FIELD_TEST_REPORT.md)
+- [Release Notes](docs/release/v0.1.0/release-notes.md)
+- [Changelog](CHANGELOG.md)
+- [Docs Index](docs/README.md)
+- [PRD: Why](docs/design/prd/01-why.md)
+- [Architecture](docs/design/prd/02-architecture.md)
+- [WBS v0.1.0](docs/wbs/v0.1.0/wbs-v0.1.0-index.md)
+
+---
+
+## License
+
+MIT — see [LICENSE](LICENSE).

cauterule-0.1.0/pyproject.toml

@@ -0,0 +1,132 @@
+[build-system]
+requires = ["setuptools>=64"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "cauterule"
+version = "0.1.0"
+description = "Automated standing-rule extraction from agent failures — extract, test, promote."
+readme = "README.md"
+license = {text = "MIT"}
+authors = [{name = "Debashish Ghosal"}]
+maintainers = [{name = "Debashish Ghosal"}]
+requires-python = ">=3.11"
+keywords = ["agent", "llm", "rules", "learning", "memory", "automation"]
+classifiers = [
+    "Development Status :: 3 - Alpha",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: MIT License",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+    "Typing :: Typed",
+]
+dependencies = [
+    "click>=8.1",
+    "pyyaml>=6.0",
+    "textual>=1.0",
+    "mcp>=1.0,<2",
+]
+
+[project.optional-dependencies]
+dev = [
+    "pytest>=8.0",
+    "pytest-cov>=5.0",
+    "ruff>=0.8",
+    "mypy>=1.13",
+    "types-PyYAML>=6.0",
+    "build>=1.0",
+    "twine>=4.0",
+]
+llm = [
+    "openai>=1.0",
+    "anthropic>=0.25",
+]
+all = [
+    "openai>=1.0",
+    "anthropic>=0.25",
+]
+
+[project.urls]
+Homepage = "https://github.com/deghosal-2026/CauterRule"
+Repository = "https://github.com/deghosal-2026/CauterRule.git"
+Issues = "https://github.com/deghosal-2026/CauterRule/issues"
+
+[project.scripts]
+cauterule = "cauterule.cli.app:main"
+
+[tool.setuptools.packages.find]
+where = ["src"]
+
+[tool.ruff]
+line-length = 100
+target-version = "py311"
+src = ["src", "tests"]
+
+[tool.ruff.lint]
+select = ["E", "F", "W", "I", "N", "UP", "B", "C4", "SIM", "RET", "ARG", "PTH", "TRY", "ANN", "D", "RUF"]
+ignore = []
+
+[tool.ruff.lint.pydocstyle]
+convention = "google"
+
+[tool.ruff.lint.per-file-ignores]
+"tests/**" = ["D", "ANN", "S101", "ARG", "PLR2004", "TRY003", "E501", "RUF043"]
+"src/cauterule/models/**" = ["D105", "TRY003"]
+"src/cauterule/config.py" = ["TRY003", "TRY004", "E501"]
+"src/cauterule/serialization/**" = ["TRY003", "TRY004"]
+"src/cauterule/llm/**" = ["TRY003", "E501", "ANN401", "D"]
+"src/cauterule/capture/**" = ["ANN401", "D", "TRY003", "TRY004", "TRY300", "UP", "SIM", "I", "RUF"]
+"src/cauterule/adapter/**" = ["ANN401", "D", "D301", "UP035", "TRY300", "TRY003", "E501"]
+"src/cauterule/redaction/**" = ["ANN401", "D", "E501", "SIM", "TRY003", "TRY004"]
+"src/cauterule/extraction/**" = ["ANN401", "D", "TRY003", "TRY004", "TRY300", "SIM", "E501", "B007", "I001", "ARG001"]
+"src/cauterule/replay/**" = ["ANN401", "D", "SIM", "RUF", "ARG001", "TRY003", "E501"]
+"src/cauterule/linter/**" = ["ANN401", "D", "TRY003", "E501", "W292"]
+"src/cauterule/promotion/**" = ["ANN401", "D", "TRY003", "E501", "W292"]
+"src/cauterule/packs/**" = ["D105", "TRY003", "TRY004", "ANN401"]
+"src/cauterule/injection/**" = ["ANN401", "D", "TRY003", "E501"]
+"src/cauterule/observe/**" = ["D", "TRY003", "E501"]
+"src/cauterule/loop/**" = ["ANN401", "D", "TRY003", "E501"]
+"src/cauterule/corpus/**" = ["D105", "D107", "TRY003", "E501", "ANN401", "ARG001"]
+"src/cauterule/cli/**" = ["D"]
+"src/cauterule/benchmark/**" = ["D", "TRY003", "E501", "W292"]
+"src/cauterule/integrations/**" = ["D", "TRY003", "E501"]
+"src/cauterule/tui/**" = ["D"]
+
+[tool.mypy]
+strict = true
+python_version = "3.11"
+files = ["src", "tests"]
+ignore_missing_imports = true
+disallow_untyped_defs = true
+warn_return_any = true
+warn_unused_configs = true
+
+[[tool.mypy.overrides]]
+module = ["yaml"]
+ignore_missing_imports = true
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+addopts = "-q --strict-config --strict-markers"
+markers = [
+    "slow: marks tests as slow",
+    "docker: marks tests that require Docker",
+]
+
+[tool.coverage.run]
+source = ["cauterule"]
+branch = true
+omit = ["*/tui/*"]
+
+[tool.coverage.report]
+fail_under = 95
+show_missing = true
+skip_covered = false
+exclude_also = [
+    "if TYPE_CHECKING:",
+    "raise NotImplementedError",
+    "@abstractmethod",
+]

cauterule-0.1.0/setup.cfg

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

cauterule-0.1.0/src/cauterule/__init__.py

@@ -0,0 +1,10 @@
+"""CauterRule — Automated standing-rule extraction from agent failures."""
+
+from importlib.metadata import version as _version
+
+try:
+    __version__ = _version("cauterule")
+except Exception:
+    __version__ = "0.1.0"
+
+__all__ = ["__version__"]