PROJECT FILE / DEVIL’S SILENCE: RESURRECTED
A classic MUD, resurrected from its own disassembly.
And the engineering machine built to bring it back without guessing.
> STATUS
Disasm-faithful core | deliberate design log | human-directed agent workflow
01 / THE STORY
Archaeology, design, and a method built for a codebase larger than one person can hold.
Devil’s Silence: Resurrected is a ground-up restoration of a classic early-2000s Diku/ROM lineage MUD. The original binaries survived. Much of the pristine source did not. That makes the compiled server the record: mechanics are recovered from real control flow, then rebuilt into a modern, maintainable engine.
That fidelity is only half the job. A living world still needs clear design decisions: class and lineage skill kits, aura tuning, mana costs, level gates, combat pacing, and new-character onboarding are documented with their rationale so the game can grow without losing its soul.
FOR THE ONES WHO REMEMBER
This is why your muscle memory still has a place here: known behavior is canonical, not “improved” by guesswork.
02 / TWO HALVES OF THE SAME PROJECT
The Game and The Machine
TRACK A / THE GAME
Faithful core. Living world.
The restoration protects the exact behavior players remember, then makes room for deliberate decisions where the record ends.
01 / DISASSEMBLY-FAITHFUL REVERSE ENGINEERING
FOR WHAT: Recover aura magnitudes, mana costs, skill gates, class kits, and combat formulas from the binary.
WHY: Returning players should recognize the feel in their hands, not just the name on the server.
02 / LIVING GAME DESIGN
FOR WHAT: Make logged decisions for class and lineage skills, buff tuning, and onboarding.
WHY: A restoration can evolve without becoming a rewrite of its own identity.
TRACK B / THE MACHINE
Parallelism without chaos.
A solo human lead directs specialized AI agents through hard ownership boundaries, shared project memory, and live-system feedback.
03 / THREE-AGENT ORCHESTRA
FOR WHAT: Separate engine patches, autonomous QA bots, and process coordination.
WHY: Each role stays in its lane, so the core C source never regresses from crossed wires.
04 / PROJECT-KNOWLEDGE MCP
FOR WHAT: Query the current build, bugs, design rationale, ideas, functions, areas, vnums, or unfinished commands in one call.
WHY: Institutional memory survives context windows. The project remembers itself.
03 / OPERATIONAL PROOF
The system is part of the project.
05 / LIVE QA
Bots play the game.
Named bots shop, fight, die, loot corpses, and re-equip. That is how subtle live-state failures surface: a bot that cannot pick up its gear is a better test than a green unit test.
06 / ONE TRUTH
Canonical vs. deploy.
The local workspace is source of truth. Git is a deploy satellite. The VPS is publish-only, never edited directly. Build-tree discipline keeps the right source path compiling every time.
07 / LIVE HARDENING
A live server, honestly observed.
A watchdog restarts a dead server while an isolated test port protects live players. The known limitation is named plainly: a hung-but-listening process can evade a basic watchdog.
04 / CHALLENGES & TRIUMPHS
Pattern matches lie. Control flow tells the truth.
Reconstructing behavior from a binary demands verification discipline: a grep or objdump match is not a fact until it holds against actual control flow. Parallel work across a large C codebase only works because engine changes, bots, and coordination have hard boundaries.
The live system exposes the details that plans miss: zero-damage DoT ticks, a descriptor leak that wedges login, and spawn or equip ordering failures. The triumph is a project that documents, tests, and remembers enough of itself to compound the work of a solo builder.
05 / POST-SNAPSHOT ROADMAP
The restoration becomes a platform.
The next direction is an in-engine Lua runtime, starting with MPROG-style scripting. The goal is not to bypass the faithful core; it is to make the world programmable without recompiling it, so new rooms, behaviors, and stories can continue to grow.
> CREDIT
Human lead directs the work. Specialized AI agents are disciplined tools: bounded, briefed, verified, and accountable to the project’s source of truth.