dendrite
dendrite
Biomimetic debug methodology — 7-phase system for finding root causes through Bayesian elimination, information-theoretic test selection, and cross-module immunity.
DENDRITE
Debug Engine for Natural Discovery, Root Isolation & Trace Elimination. Biomimetic 7-phase methodology that finds root causes with mathematical precision.
Install
myclaude install dendrite
Usage
/dendrite [describe the bug or paste the error]
Examples
/dendrite NullPointerException at parser.java:142 after config migration
/dendrite Tests pass locally but fail in CI — flaky, non-deterministic
/dendrite Users report intermittent 500 errors on checkout endpoint
/dendrite --mode=deep Production database queries suddenly 10x slower after deploy
Modes
| Mode | Use For | Phases |
|---|---|---|
quick | Known patterns, obvious causes | 1→2→5→6 |
standard | Most bugs (default) | All 7 |
deep | Critical bugs, production impact | All 7 expanded |
probe | Emergent bugs that resist analysis | Experimental |
How It Works
DENDRITE operates in 7 phases inspired by the immune system's response cycle:
- PRESERVE — Snapshot state before any intervention (forensic preservation)
- CLASSIFY — Triage the bug's domain (Clear/Complicated/Complex/Chaotic)
- COLLECT — Gather evidence at minimum invasiveness (Heisenberg protocol)
- DIAGNOSE — Bayesian elimination tree with Information Gain test selection
- COMPREHEND — Causal narrative: why-did / why-didn't (not just "where")
- PROPAGATE — Scan for siblings across the codebase (species, not individual)
- EVOLVE — Register pattern, measure MTTRC, strengthen the system
What Makes It Different
- Information Gain Engine — Mathematically selects the optimal next test (Shannon 1948)
- Cynefin Triage — Different strategies for different problem domains
- Comprehension Phase — Inspired by Ko's Whyline (8x debug speedup)
- Cross-Module Immunity — Finds bug siblings via dependency graph scanning
- Evidence Clustering — Prevents overconfidence from correlated evidence
- Antifragile Evolution — System gets stronger with each debug cycle
Requirements
- Claude Code >= 1.0.0
- Access to source code and logs of the system being debugged
- At least one observable symptom (error, crash, wrong behavior)
Academic Foundations
Grounded in 15 validated sources including Shannon (Information Theory), Ko & Myers (Whyline), Parnin & Orso (ISSTA Impact Award), Zeller (Delta Debugging), Klein (RPD), and Vesely (Fault Tree Analysis). See references/academic-foundations.md for full bibliography.
<sub>Built with MyClaude Studio Engine</sub>
Reviews (0)
Loading reviews...