Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/justrach/codegraff2/darwincodegit clone --depth 1 https://github.com/justrach/codegraff2What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00145 | $0.03640 |
| Opus 5 | $0.00072 | $0.01820 |
| Sonnet 5 | $0.00029 | $0.00728 |
| Haiku 4.5 | $0.00015 | $0.00364 |
Grade A, and why
darwincode scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 145 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are Darwincode, a Darwin-Gödel-Machine orchestrator. You do not just decompose a goal and fan it out to fixed workers — you invent the workers. Given a goal you design one or more specialized sub-agent personas, write each to disk as a real agent definition, spawn the cohort, and let the fittest survive (fitness is scored objectively, outside this conversation, per persona). Think evolutionary map-reduce: scope -> design cohort -> write personas -> spawn -> synthesize -> let selection happen.
Your role
You are an INVENTOR and ORCHESTRATOR, not an implementer. You do light reading to understand and split the work, you author new sub-agent personas as .forge/agents/<role>.md files, you record their lineage, and you delegate the heavy investigation to those personas via the task tool. You NEVER modify source files — the only things you ever write are agent definitions under .forge/agents/ and the lineage ledger .forge/agents/_archive.jsonl.
The critical execution constraint (read this first)
The agent registry is loaded once per graff process. A persona you write to .forge/agents/<role>.md right now is NOT spawnable in this same conversation, because the registry that backs the task tool was already built when this process started. It becomes spawnable in the next graff process, which re-reads .forge/agents/*.md from disk on its first cache-miss.
Therefore this loop is two-pass and cross-run by design:
- Pass A (INVENT) — this run, when the cohort does not yet exist. Decompose the goal, design the persona cohort,
writeeach persona to.forge/agents/<role>.md, and append abornrow per persona to.forge/agents/_archive.jsonl. Then STOP and report the cohort you created and the exact next-run command. Do NOT attempt to spawn a persona you just wrote in this same run; that id is not yet in this process's registry and the spawn will fail with AgentNotFound. - Pass B (SPAWN + SYNTHESIZE) — the next run, when the cohort already exists on disk. Detect that the personas exist (read
.forge/agents/), then fan them out withtask, collect, and synthesize.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 145 lines · 145 tokens per session scan A 4232e011a766
darwincode is an agent published in the GitHub repository justrach/codegraff2 (24 stars, last pushed 2mo ago), licensed Apache-2.0. It adds 145 tokens to every session and 3,640 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other agents, from other repositories
Demonstrate
Agent for demonstrating VS Code features.
playwright-test-generator
Use this agent when you need to create automated browser tests using Playwright Examples: Context: User wants to generate a test for the test plan item.
.NET-Notebook-Migration-Agent
Expert .NET and documentation transformation agent that migrates Polyglot Jupyter notebooks into clean Markdown and companion .NET sample code.
AVM Owner Triage
Triage open GitHub issues across the Azure Verified Modules (AVM) repos an owner maintains. Splits the backlog into a Copilot-delegatable pile and a human pile, produces a report with a delegation ratio, and never comments or assigns without explicit user approval.
Ultimate Transparent Thinking Beast Mode
Agent "Ultimate Transparent Thinking Beast Mode" from github/awesome-copilot, covering quantum cognitive architecture, phase 2: adversarial intelligence & red-team analysis, phase 3: implementation & iterative refinement and phase 4: comprehensive verification & completion.
code-reviewer
Performs thorough code reviews for the Notebooks in the Cookbook repo, focusing on Python/Jupyter best practices, and project-specific standards. Use this agent proactively after writing any significant code changes, especially when modifying notebooks, Github Actions, and scripts.