Iteration loop

Single-file fix

1. Symptom. "Players see double mileage in the speedometer."
2. Agent calls read_file({ resource: "speed_hud", path: "client/main.lua" }).
3. Agent calls write_file({ resource: "speed_hud", path: "client/main.lua", content: "<patched>" }).
4. Agent calls restart_resource({ name: "speed_hud" }).
5. Agent calls tail_console({ lines: 20 }) to verify no errors on start.

You confirm in-game; done.

UI iteration

1. Agent edits web/src/App.vue.
2. Agent calls run_shell({ resource: "my_res", command: "npm", args: ["run", "build"], cwd: "web", timeoutMs: 60000 }).
3. Agent calls restart_resource({ name: "my_res" }).
4. You reopen the UI (F6 / /myres).
5. Agent calls screenshot_nui({ resource: "my_res" }).
6. Agent reads the PNG, describes what it sees, proposes the next tweak.

Average loop time on a dev box: ~5–10 seconds per change.

Multi-resource change

1. Agent picks a name and runs scaffold_fivem_resource_workflow (mandatory grill).
2. After your confirmation, agent: create_resource, refresh_resources, run_shell for any UI scaffolding, then a batch of write_file calls.
3. ensure_resource returns stateAfter === "started" and the post-start console lines.
4. If start failed, agent calls tail_console to find the syntax error and offers a single-file fix before re-writing.

Live player workflow

1. You: /agent_test_optin.
2. Agent: register_test_subject({ serverId: 1 }).
3. Agent: get_player_state({ serverId: 1 }) to see coords, ped, health.
4. Agent: esx_call_player({ serverId: 1, method: "getInventory" }) to see items.
5. Agent: client_call_native({ serverId: 1, native: "SetEntityCoords", args: ["$ped", 200, -800, 30, false, false, false, true] }) to test a teleport.
6. Agent: client_call_native({ serverId: 1, native: "GetEntityCoords", args: ["$ped", true] }) to verify the move.

What goes wrong + how the agent recovers