Jamon Holmgren jamonholmgren
Status: DRAFT (2026-06-28) — investigation just started. No implementation yet.
Slug: combat-perf-npc-gun-scan
Task source: Interactive human request (Jamon, 2026-06-28). After the tree-LOS native win (~90× faster query), hunt for the next perf opportunities. Jamon profiled a live combat mission with the F4 Perf panel and shared screenshots; this worksheet captures the read and the investigation.
Summary: Orchestrator protocol for driving multi-agent task execution. One coordinating agent (the orchestrator, often Codex) slices a task into parallel-safe units and dispatches worker agents (often Claude) via the Tools/agent_review CLI wrapper. Covers work slicing, context hygiene, temp report files, parallelism, avoiding conflicting edits, polling/tailing results, and output validation. Use when the user asks an agent to act as orchestrator/coordinator/dispatcher rather than do the work directly.
Key Files: Tools/agent_review, AGENT_WORKFLOW.md, AGENT_REVIEW.md, CODING_CONVENTIONS.md
Status: IMPLEMENTED
I've reached a pretty solid workflow for getting the most from agents in my code base. This setup and workflow has significantly reduced how often I have to fix obvious issues caused by agents doing the wrong thing.
This entire guide was hand-written by me, Jamon, with no AI-generated text nor any AI-suggested text or review. Fully human-generated.
My philosophy is based on these concepts:
Lessons learned on how to use AI agents effectively. In no particular order. Edited from time to time.
With agents, there should be no such thing as “tech debt”. An agent should simply pay down every bit of tech debt before presenting you with the “finished work”. Unlike humans, agent time is not very valuable; it can continue to work on something until it’s done, and shouldn’t make concessions assuming human constraints.
If an agent tells you that it’s “leaving something for later”, tell it to go finish it first before saying it’s done.
Per: https://x.com/jamonholmgren/status/2043385696535331324
mkdir -p "$HOME/homebrew"
git clone https://github.com/Homebrew/brew "$HOME/homebrew/.linuxbrew/Homebrew"
mkdir -p "$HOME/homebrew/bin"
ln -s "$HOME/homebrew/.linuxbrew/Homebrew/bin/brew" "$HOME/homebrew/bin/brew"
# Homebrew (per-user)
| #!/usr/bin/env bash | |
| set -euo pipefail | |
| # Downloads the latest Terrain3D build artifacts from the main branch | |
| # and installs them into addons/terrain_3d/. | |
| # Can be run from the project root or any subfolder. | |
| REPO="TokisanGames/Terrain3D" | |
| ARTIFACT_NAME="Terrain3D" |
Determine your mode from the user's request before doing anything else:
- Plan — User describes a feature, bug, or idea. The Project Manager reads GAME_DESIGN + PROGRESS + TODOS, consults relevant experts to flesh out the plan, produces concrete steps, creates/updates TODO entries, and stops for user review. No code yet.
- Build — User says "do this" or points at a specific TODO. The Project Manager identifies which experts and docs are relevant, gets their input on approach, then implements. After implementation, experts review and the PM runs the quality checklist. Lint and commit.
- Review — User asks for review, or you're finishing a build. The Project Manager runs relevant experts against recent changes and produces a structured report with an overall assessment.
- Loop — User explicitly requests the agent loop. Read Docs/AGENT_LOOP.md and follow it.
(as of March 20, 2024)
Yamaha AG03 3-Channel Mixer / 1 Microphone $150
Electro-Voice RE320 Large Diaphragm Dynamic Vocal Microphone $200 used, $300 new.