AI-native software engineer | I build with agents and review every change before it ships

I split the queue into focused tasks and give each one its own execution slot. One agent can implement while another checks the diff, runs the relevant tests, or tries to break the result. Every change still needs a clear scope, reviewable commit, test evidence, a PR, and confirmation that the merge really reached the remote branch.

This lets me handle more work without treating review and verification as optional.

An agent that stays until the work is actually done

Most coding agents hand back a patch and stop. I built Simplicio Loop to keep the work moving until the tests pass, the review is clean, and the change is merged.

discover → plan → implement → test → verify → review → merge → close → watch 24/7

It works with any LLM or runtime. The loop keeps receipts, controls risky actions, watches cost, and continues from the last verified state instead of starting over.

These are published project-level maximums, not a promise that every workload will hit the same number. I also kept the original comparison: Caveman 65%, RTK 80%, and Simplicio at up to 96%.

I built these projects because the same problems kept coming back: agents losing context, tools editing the wrong thing, tasks stopping before review, and results with no proof behind them.

simplicio-agent · v2026.7.20 Problem: autonomous agents can act without durable checkpoints or a trustworthy audit trail. What I built: gated actions, checkpoints, evidence receipts, MCP, skills, multi-model freedom, and Rust-backed deterministic boundaries.

simplicio · v3.5.2 Problem: AI coding tools waste context and fragment chat, mapping, editing, and multi-agent work. What I built: one coding-agent runtime and distribution surface, preserving the published up to 96% token-savings result.

simplicio-dev-cli · v0.16.1 Problem: a one-line coding request is not a delivery process. What I built: mapped context, deterministic edits, tests, retries, reviewable diffs, and evidence. The project is positioned around 99% accuracy across major LLM hosts.

simplicio-mapper · v0.23.1 Problem: agents begin coding blind when repository structure, dependencies, symbols, and precedent are missing. What I built: a stack-neutral project map and compressed context pack available from minute one.

simplicio-prompt · v1.14.1 Problem: large agent systems waste context searching for capabilities and memory. What I built: yool + tuple-space + HAMT addressing, guardrails, receipts, and a published 75% token-economy position.

simplicio-sprint · v1.2.14 Problem: sprint tickets do not carry repository architecture or delivery proof. What I built: Jira/Azure DevOps intake, repository mapping, multi-agent dispatch, and result verification.

simplicio-local Problem: cloud inference adds latency, recurring cost, network dependency, and privacy constraints. What I built: 100% on-device Apple Silicon inference paths using MLX, Metal, and ANE-oriented architecture.

simplicio-loop-marketing · v0.4.0 Problem: marketing teams are locked into one provider and disconnected manual tools. What I built: brief → script → creative → caption → compliance → publish → metrics → ads, with provider routing through configuration.

PiAPI-Skills · v1.2.0 Problem: media-generation capabilities are fragmented across agent platforms. What I built: one portable skill surface for Claude, Codex, Hermes, OpenClaw, Cursor, Windsurf, and generic agents.

WaveSpeedAI-Skills · v1.2.0 Problem: teams repeatedly rebuild provider integrations for every agent host. What I built: one installer and CLI exposing 700+ models to agentskills.io-compatible environments.

flowchart LR
    M[Mapper<br/>repository context] --> D[Dev CLI / Code<br/>verified changes]
    D --> R[Runtime<br/>policy + effects]
    R --> A[Agent<br/>tools + checkpoints]
    A --> L[Loop<br/>orchestrate to completion]
    L --> E[Evidence<br/>review + merge + watch]
    L --> MK[Marketing<br/>specialized pipeline]
    W[PiAPI + WaveSpeed<br/>portable skills] --> A

Top public repositories by stars, excluding forks and the profile repository. Snapshot values are preserved; badges and graphs below update live.

Star history · current top projects

Contribution activity

Ranking refreshed from the public GitHub API on 2026-07-22. Live cards and badges may move after the snapshot. Clone counts and release-download counts are not embedded because GitHub's Traffic API requires authenticated access.

GitHub records 38 commit contributions from me to the original Hermes Agent in 2026 through July 22. The work landed in parts of the product people actually touch:

• Fixed dashboard and terminal details that got in the way, including provider search, mobile scrolling, and selection/copy behavior.

• Made Hermes less fragile in Docker by improving gateway detection, environment propagation, cleanup of orphaned processes, and recovery from temporary failures.

• Added compatibility fixes for fallback API keys, DeepSeek V4 thinking blocks, provider profiles, browser detection, and IMAP.

• Tightened Kanban migrations and dependency handling, CJK session search, checkpoints, and optional dependency packaging.

• hermes-turbo-agent: performance branch with benchmarks, visual comparisons, token reporting, and safe hot-path research.

• x-bookmarks-panel: turns saved X posts into an actionable local-first AI queue.

• x-virality-skills: source-grounded workflows for X's For You algorithm.

• sistema-sindico: PHP/MySQL condominium management with a mobile-ready API foundation.

• Brazil banking CLI suite: Open Finance and banking experiments across BB, BTG, Inter, Matera, PagBank, PicPay, and Open Finance BR.

I like AI when it survives contact with a real repository. It should understand the codebase, remember why a decision was made, coordinate work safely, and leave enough evidence for someone else to trust the result.

I am building the tooling I wanted for my own work: less context lost between sessions, less repetition, and more changes that make it all the way from an issue to a verified merge.

Jesse Daniel Brown, PhD has been an important mentor to me. He is from California and has written more than 100 scientific articles. His view of programming and AI is grounded in education and humanitarian work, and it has shaped how I think about building technology that is useful to people.

• GitHub: @wesleysimplicio
• X: @wesleysimplic
• Instagram: @wesleysimplicio
• LinkedIn: wesleysimplicio
• YouTube: @wesleysimplicio