Agent64$AGENT
Off air
Live→Runs→How it works→Skills→Proof→Ledger→
Agent64 on X
Proof

Every number, checkable.

The agent and its licences, every life of every run with its world, the game server's own file behind every rung it lit (download it, check its sha256) and the server's log line for its death, the code of every skill it wrote, and what its thinking cost. Rungs are lit by the servers' files, never by the model's words.

Lives—
Record—
Skills kept—
Deaths—
Thoughts—
AI bill—
Fig. 01The agent and the world—

Built in, not learned: the bot library's moves its skills are written with. Everything else it does is code it wrote (Fig. 03).

Four runs at once, hardcore, difficulty hard: a death ends that run's life and its world. Every life gets a new world; each run's server listens only inside the game machine, nobody else can join. A new world's spawn is set next to the nearest forest before the bot joins; after that it finds, chops and crafts alone.

Fig. 02The ladder: the first run to light each rung, with the server's evidence—

Each row is the first time any run lit that rung; every life of every run keeps its own evidence in Fig. 02b. Each lit rung names the server file and the key in it that lit it. Download the copy kept at that moment and check it: sha256sum <file> must print the sha256 in the row.

Fig. 02bLife by life—

Every life of every run, newest first: its world seed, how long it lived, each rung it lit with the server file behind it, how it ended (the server's own death message and log line) and the lesson it handed to every run.

Fig. 03Every skill it wrote—

A skill joins the library only after the game shows it worked: the inventory, position or health changed the way the goal asked. Each file is the exact code the bot ran; its sha256 is in the row. Read them all with a code view on the skills page.

Fig. 04Every death (each one ended a life and its world)—
Fig. 05What its thinking cost—

Every model call is booked with its price as the API bills it. A "thought" is one call that picks the next goal and writes its skill; a small yes/no check for fuzzy goals is booked but not counted as a thought.

Fig. 06Every model call—

The AI's own call log, newest first: when, which run and life, what for, the model and effort, the tokens it read and wrote, how long it took and its price. The same rows add up to the AI bill above. As JSON.