How do I install the Debugging Reservoir skill?

Run npx skillmds@latest add d-o-hub/debugging-reservoir in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Debugging Reservoir safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Debugging Reservoir?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Debugging Reservoir free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Debugging Reservoir?

d-o-hub (@d-o-hub) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Debugging Reservoir by d-o-hub — Agent Skill

Name: Debugging Reservoir
Author: d-o-hub

File contents

Debugging the Reservoir

Architecture Overview

The reservoir is a sparse Echo State Network in src/reservoir.rs:

Reservoir: core ESN with sparse adjacency-list weights
ChaoticReservoir: wrapper adding noise perturbation for edge-of-chaos dynamics

Sparse Weight Format

Weights are Vec<Vec<(usize, f32)>> — each row is a list of (column_index, weight).

Input weights (w_in): 32 connections per neuron
Reservoir weights (w_res): 64 connections per neuron (or min(64, size))
Dot product via dot_sparse_row(&weights[i], &values)

Spectral Radius Debugging

The spectral radius controls chaos level. Must be in [0.9, 1.1].

Symptoms of wrong radius

Symptom	Likely Cause
State saturates to all ±1	Radius too high (> 1.1)
State decays to all ~0	Radius too low (< 0.9)
Non-deterministic test failures	Using `new()` instead of `new_seeded()`
Different results across runs	Unseeded RNG

How radius is estimated

Power iteration (16 steps) in estimate_spectral_radius():

Start with uniform vector v
Repeat: v = W * v, then normalize
Compute Rayleigh quotient: v^T W v / v^T v

If radius is wrong after init, check:

Are weights being scaled? (scale = target / current)
Is current_radius returning 0? (degenerate weights)
Is degree too low? (< 8 may give poor estimates)

to_hypervector() Projection

Projects reservoir state → HVec10240. Requires size >= 10240.

Divides state into 10240 chunks
Each chunk sum > 0 → bit = 1

Common errors

InvalidDimension { expected: 10240, actual: N } — reservoir too small
All-zero hypervector — reservoir state hasn't been activated (call step() first)

Chaotic Reservoir Noise

ChaoticReservoir adds uniform noise ±chaos_strength to each input element. Uses a separate RNG seeded with seed ^ 0xA5A5_5A5A_F0F0_0F0F for independence.

Diagnostic Checklist

Is step() being called before to_hypervector()?
Is the reservoir reset() between independent sequences?
Is size >= 10240 for projection?
Is spectral radius in [0.9, 1.1] after construction?
Are tests using new_seeded(..., 42) for determinism?

---name: debugging-reservoirdescription: Debug and tune the echo state network reservoir. Use when diagnosing spectral radius issues, chaotic dynamics problems, sparse weight anomalies, or reservoir-to-hypervector projection failures.---# Debugging the Reservoir## Architecture OverviewThe reservoir is a sparse Echo State Network in `src/reservoir.rs`:- `Reservoir`: core ESN with sparse adjacency-list weights- `ChaoticReservoir`: wrapper adding noise perturbation for edge-of-chaos dynamics## Sparse Weight FormatWeights are `Vec<Vec<(usize, f32)>>` — each row is a list of `(column_index, weight)`.- Input weights (`w_in`): 32 connections per neuron- Reservoir weights (`w_res`): 64 connections per neuron (or `min(64, size)`)- Dot product via `dot_sparse_row(&weights[i], &values)`## Spectral Radius DebuggingThe spectral radius controls chaos level. Must be in `[0.9, 1.1]`.### Symptoms of wrong radius| Symptom | Likely Cause ||---|---|| State saturates to all ±1 | Radius too high (> 1.1) || State decays to all ~0 | Radius too low (< 0.9) || Non-deterministic test failures | Using `new()` instead of `new_seeded()` || Different results across runs | Unseeded RNG |### How radius is estimatedPower iteration (16 steps) in `estimate_spectral_radius()`:1. Start with uniform vector `v`2. Repeat: `v = W * v`, then normalize3. Compute Rayleigh quotient: `v^T W v / v^T v`If radius is wrong after init, check:- Are weights being scaled? (`scale = target / current`)- Is `current_radius` returning 0? (degenerate weights)- Is degree too low? (< 8 may give poor estimates)## to_hypervector() ProjectionProjects reservoir state → `HVec10240`. Requires `size >= 10240`.- Divides state into 10240 chunks- Each chunk sum > 0 → bit = 1### Common errors- `InvalidDimension { expected: 10240, actual: N }` — reservoir too small- All-zero hypervector — reservoir state hasn't been activated (call `step()` first)## Chaotic Reservoir Noise`ChaoticReservoir` adds uniform noise `±chaos_strength` to each input element.Uses a separate RNG seeded with `seed ^ 0xA5A5_5A5A_F0F0_0F0F` for independence.## Diagnostic Checklist1. Is `step()` being called before `to_hypervector()`?2. Is the reservoir `reset()` between independent sequences?3. Is `size >= 10240` for projection?4. Is spectral radius in `[0.9, 1.1]` after construction?5. Are tests using `new_seeded(..., 42)` for determinism?

Frequently asked questions

Debugging Reservoir