Skip to content

CauterRule v0.1.0

Latest
Latest

Choose a tag to compare

View all tags
@deghosal-2026 deghosal-2026 released this 06 Sep 05:21
· 1 commit to main since this release
v0.1.0
9dffb8d

CauterRule v0.1.0 Release Notes

Release date: September 6, 2026
Tag: v0.1.0
Milestone: M31 — Release Readiness

What is CauterRule?

CauterRule is an open-source sidecar that turns repeated agent failures into permanent standing rules. It captures failure trajectories, extracts structured when X, do Y rules using LLMs, replay-tests them against historical scenarios, and promotes only rules that survive — with linter checks, conflict detection, and git-based provenance.

Think of it as CI/CD for agent behavior: test a rule against the regression suite before merging it in.

What's New in v0.1.0

Core Loop

  • Trajectory capture with automatic secret redaction
  • Multi-pass LLM extraction (3x temperature variation, draft tournament ranking)
  • Historical replay engine with deterministic evidence reports
  • Promotion gate with 6 deterministic checks (sample floor, effect size, confidence, frozen sections, edit distance, drift)
  • Rule linter detecting vagueness, tautologies, duplicates, contradictions, untestable and unsafe directives
  • Versioned YAML rule store with git provenance, rollback, and archive

Developer Experience

  • 25+ CLI commands including demo, extract, test, promote, inject, rewind, counterfactual, story, mcp
  • 7 export formats: .cursorrules, CLAUDE.md, AGENTS.md, .windsurfrules, aider.conf.yml, markdown, JSON
  • MCP server with 4 tools for any MCP-compatible agent
  • @cauterule.watch decorator for any Python agent — no framework lock-in
  • Bundled pack-git with 10+ pre-built git rules for zero cold-start
  • Configuration via cauterule.toml with environment variable overrides

Corpus & Benchmarks

  • 13 corpus types totaling 394 trajectories across curated and raw sets
  • Golden trajectory set with known expected rules for regression testing
  • Safety corpora: successes, failures/negative, nearmiss, noisy, corrections
  • Raw corpora: OpenCode sessions, synthetic scenarios, CI failure logs, sibling-repo agent runs, cross-session repeats
  • Field test runner supporting local OMLX and cloud OpenRouter LLMs

Field Test Results

The v0.1.0 field test evaluated 4 models across 13 corpus types:

Model Type Total 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%
  • Cloud models achieved near-perfect parse reliability across all corpora
  • meta-llama/llama-3.1-8b-instruct was the strongest cost-effective model tested
  • Safety corpora remain the hardest unsolved area across all models
  • The remaining bottleneck is replay quality and rule specificity, not parsing

Full report: docs/field-test/v0.1.0/FIELD_TEST_REPORT.md

Security

  • Production dependencies audited clean (click, pyyaml, textual, mcp)
  • truffleHog scan: all findings are test fixtures (intentional fake tokens in redaction tests)
  • Redaction engine strips secrets before LLM extraction and export

Known Limitations

  • TUI review (cauterule review) deferred to v0.2.0
  • Formal observability subsystem deferred to v0.2.0
  • Adversarial corpus generation deferred to v0.2.0
  • Distribution channels (Homebrew, standalone binary, GitHub Action) deferred to v0.2.0
  • Replay matcher uses heuristic substring + token overlap; semantic matching planned for v0.6.0

Installation

pip install cauterule

Quick Start

cauterule demo          # seeded failures, full loop in 60s
cauterule init          # scaffold a new project
cauterule extract <trajectory.jsonl>   # extract a candidate rule
cauterule test <rule>   # replay-test against history
cauterule promote <rule> # promote to permanent store

Documentation

Assets 2
Loading