Coding

Decompile an old game to readable C++ with a fleet of agents

  • Level Intermediate
  • Setup1–3 hrs (estimated)
  • Evidence Some evidence
  • CostNot stated in source
  • No-codeNo
Unnamed agentSource-linked

What they did

Over 3 months and 500+ billion tokens, a developer and community members had AI agents decompile a popular first-person shooter into readable C++ that compiles, and wrote up how they set up the harness and orchestration.

How it went

Long, detailed write-up; 500B+ tokens. Only decompile software you have rights to

Worth knowing

  • Long, detailed write-up; 500B+ tokens. Only decompile software you have rights to.

Prompt

Cleaned version rewritten from the source — not a direct quote.

For [function address range] in [binary I own], decompile each function to readable C++ that compiles and matches the original behavior. Write a test that compares outputs, mark anything uncertain, and keep a log of which functions are done.

Recommended

Try it in Grok Bot

  1. Copy the prompt (button below).
  2. Open Grok Bot chat and paste it. Replace any [bracketed] placeholders.
  3. Connect the listed tools (as needed) as connectors when Grok Bot asks.
  4. If the task repeats, ask: "Make this a daily routine."

Alternatives: the same prompt works in Claude, ChatGPT Agent, OpenClaw.

Set it up

Stack: Coding agents (orchestrated fleet)

  1. Read the original source for full context.
  2. Connect the tools your agent needs.
  3. Paste the prompt and replace [bracketed] placeholders.
  4. Run once manually and check the output before scheduling it.

Generic suggested steps, not from the source.

Evidence

Long, detailed write-up; 500B+ tokens. Only decompile software you have rights to

Source: Web · Verified by us: not yet · Report broken or outdated

Related use cases

CodingX · 2026-10

Hermes Agent Nearly Doubled Speech-to-Text Accuracy on Hard Words: 99 to 183 of 240

A user turned his 15K saved Wispr Flow dictations into a test set for the local Whisper model behind his Hermes Agent. The agent turned his Wispr dictionary into a hint list, diffed raw vs corrected text to find words Whisper keeps mishearing, made find-and-fix rules, and tested every rule against the full history. With no fine-tuning, dictionary words went from 99/240 to 183/240 on Wispr clips and 19/45 to 43/45 on sentences he read aloud. It now archives the data nightly and looks for new fixes weekly.

  • Level Advanced
  • Setup1–3 hrs*
  • Evidence Strong evidence
Hermes AgentSource-linked
CodingReddit

Non-engineer built a B2B SaaS with Codex: 2 paying clients

A non-engineer built a 'boring' vertical B2B SaaS with an orchestrator chat plus worker chats. Two paying clients and 6 leads, about $800 in credits spent.

  • Level Intermediate
  • Setup1–3 hrs*
  • Evidence Some evidence
CodexSource-linkedNeeds verification