UNITED STATES — EST. 2024 · BUILDING IN PUBLIC / UTC−04:00

Practical tools
for real systems.

I'm TheElephantCoder — a developer who likes to understand how things actually work. Usually by taking them apart first. I build small, focused, open-source tools: from bootable recovery environments to security auditors and low-level Rust engines. No bloat. Just things that are useful — mostly to me, hopefully to you too.

View selected work Repositories
05 Public repos
all open source
03 Featured builds
maintained
∞ Curiosity
systems first
theelephantcoder — zsh — 80×24
$ cd theelephantcoder-profile
$ cat README.md
Hi, I'm @TheElephantCoder — tinkering with tech.
→ security · systems · docker · low-level
$ ls projects/
agent-harness/   bit-compact/   claw-auditor/
$ cat stack.txt
python   rust   c   c++   bash   docker
$
01 — APPROACH

Built to be understood,
not just to run.

I like to build tools that are transparent about what they do. Small surface area, readable code, reproducible builds. The kind of thing you can open, audit, and learn from — without needing a map and three energy drinks.

02 — SELECTED WORK

Three projects
I kinda finished

All projects are open source. No wrappers around APIs — actual tooling you can build, boot, or run locally. And yes, they work on my machine.

PERFORMANCE LAYER · AGENT HARNESS · NODE / PYTHON
22 COMMITS · 2★

agent-harness

Performance layer for coding agents — skills, instincts, memory, security and research-first.

I built this because agents kept re-reading the same 250k LOC for 3 minutes and then guessing anyway. Harness caches what matters, enforces a research step before code, and transpiles one set of skills to 8 tools — Claude Code, Codex, Opencode, Cursor, Kiro, Cline, Aider and more.

  • 13s cold start — was 180s · tokens down 65% · tool calls down 40%
  • Skills, instincts, memory (MEMORY.md), security (default deny), research-first flow
  • One harness, 8 adapters — write once, use everywhere
TypeScriptNodePythonAgentsPerformanceSecurity
View repository Docs
● ● ● harness — init
$ harness init --auto
✓ wired to opencode, claude, cursor

$ harness skill add research-first
✓ available in all 8 harnesses

$ harness bench --quick
cold start 13s · p99 hook <100ms
8 ADAPTERS · 22 COMMITS npm · pipx · brew · apt
cold start13s
tokens-65%
hooks<100ms
FEATURED · RUST · COMPRESSION ENGINE
2026 — 1★ · 15 commits

bit-compact

Zero-dependency Rust engine for SQ8 scalar quantization. 4× smaller embeddings, zero-alloc seeks.

A low-level multimodal embedding store — 32B Big-Endian header, per-dimension calibration, Send + Sync reader, 10 APIs, and a self-contained serve GUI. Built for concurrent analytical scans with cache-line and disk-block awareness.

  • 4× reduction · f32 → u8 · SQ8 linear quantization
  • 1 seek · zero-allocation random access
  • Dockerized, stripped binary · tini runtime
RustDockerVector DBQuantizationZero-dep
View repository
● ● ● storage.rs — CompactReader
// 1 seek, 0 alloc — Send + Sync
let r = CompactReader::open("vectors.btcp")?;
let mut buf = vec![0u8; r.dims()];
r.get_quantized_into(42, &mut buf)?;
let v = r.get_vector(42)?;

// header 32B BE exact
MAGIC | ver | dims | count | footer
min/max → quantized stream → sha256
10 APIS · BENCHED · VALIDATED cargo test · clippy —D warnings
header32 B
compression4×
alloczero
SECURITY · NODE.JS · OPENCLAW SKILL
1 commits · 44 RULES

claw-security-auditor

Autonomous auditor for OpenClaw skills. Static only — never executes untrusted code.

Scans installed skills for shell execution, remote code download, file deletion, exfiltration, obfuscation, and 38 more patterns. Assigns a 0–100 risk score, risk level, and generates mitigation recommendations. Dashboard at localhost:7777 with no build step.

  • H1–H16 high · M1–M20 medium · L1–L10 low + entropy (H4e)
  • CLI: --output json|markdown|csv · --fix · --compare · --trust
  • Zero dependencies, read-only, file watcher
JavaScriptNodeSecurityStatic Analysis
View repository Skill spec
RISK MODEL
0–29 LOW
optional review
30–59 MEDIUM
manual review
60+ HIGH
disable / sandbox
H2 / H4 always → HIGH · shannon entropy · 10 API dashboard
03 — STACK & FOCUS

Tools I reach
for most often.

The stuff I actually use, usually while pretending I remember all the flags.

LANGUAGES
Python C C++ JavaScript Rust Bash
SYSTEMS & PLATFORMS
Linux Docker Git Raspberry Pi Networking Security
WHAT I LIKE WORKING ON
  • Low-level system behavior & I/O
  • Security analysis tooling
  • Lightweight developer utilities
  • Hardware + software interaction
  • Learning by building — small, finished projects
04 — CONNECT

Building in public.
Follow the work.

I share code on GitHub and occasional builds elsewhere. No hiring pipeline here — just open source and notes. If something I built is useful (or breaks spectacularly), open an issue or PR.

NOTE

I'm not actively looking for contract work — I build what I find interesting and share it. It's cheaper than therapy. If you want to collaborate on tooling, security research, or a focused open-source idea, the best way is a well-scoped GitHub issue.

OPEN TO MEANINGFUL PRS & DISCUSSIONS
Browse repositories
Response time — best effort · Issues preferred over DMs

* 3 featured repos — agent-harness, bit-compact, claw-security-auditor. Excludes profile README (TheElephantCoder) and portfolio site (theelephantcoder.github.io). Total public repos: 5.