Stickbash v4 COORDINATOR CHECK-IN (every 15 min). FULLY AUTONOMOUS (the user, 24 Oct: "now do autonomously, create tracked checklist, and set rules, crons etc"; 25 Oct: "U r fully autonomous, u can take decisions. Always prioritize quality, maximize ur token usage for that, do what recommended always"). If the user writes (in chat OR in the INBOX), answer them first. Use `date` for real times. Work must NEVER stay stopped.
== STEP 0a · THE INBOX == In ~/StickBrawl-v3/v4/feedback run `bash _tools/inbox.sh pull`. Every message it prints is the USER'S OWN WORDS (from the index page's INBOX panel); act on it like a chat message and answer with `bash _tools/inbox.sh reply <id> "<answer>"` (stand-alone plain words). At the END of every tick set the status line: `bash _tools/inbox.sh status "<1-2 lines>"`.
== THE RULES (full text: _tools/58k_BRIEF.md §9) == QUALITY NEVER GOES DOWN; absolute uniqueness (first-built keeps the idea); brutal XT with real faces at XT/VH; the 16:10 LADDER STANDARD (6 different final pictures); gold FARIS BASHA plate always visible; 33 weapons (TIME WATCH and WING BLADE removed); films 25/15 + 7 skins, fists 25; mids 10/6, <=3 s, sample at K58.finalT(def,o); ZONE: the six picks rotated, push in 0.36 s, MID_HOLD 0.62, push out, outsiders keep the arena camera; mids on small maps (switch, default ON); EVERYTHING HAS A SWITCH; GATE K4.0 LIFTED: KILL_PORT.md, P0 TF-PORT-ENGINE (StickFight-v3-kills / tf-kills), P1 TF-PORT-SYSTEM (StickFight-killsys / tf-killsys), P2 SMG first after the harness passes; a ported kill merges only after the harness passes at six levels AND I've looked at the sheet; the user views pages LOCALLY via 36_index.html.
== STEP 0 · HEALTH == /Volumes/Workspace mounted (else `hdiutil attach /Volumes/MySSD/Workspace.sparsebundle`); df, swap, load; autosync alive (pid in /Volumes/Workspace/Scratch/coord/autosync.pid, else restart `nohup bash _tools/pages_autosync.sh &` and rewrite the pid) + ix_status fresh; inbox service up; orphaned headless Chromes (kill only those PIDs, rmtree profiles via python).
== STEP 1 · SLOTS — MAX 12 SUBAGENTS AT ONCE (user, 25 Oct 22:05: "Only run 12 subagent at a time not more than that") == Read _tools/coord/AGENT_SLOTS.md. Keep exactly the ACTIVE set running (resume any ACTIVE agent that stopped, with its exact next step; ALWAYS ASSUME THE SESSION LIMIT IS RESET). When an ACTIVE agent's lane is finished, move it to RESTING and resume the QUEUE head in its slot; update the file. Never more than 12 running; no agent spawns helpers/forks. NEVER resume a604f983e7d963ede, a3f6b216eca08b47a, a8ddf72b6dfb1e3aa, acc20e3f20868f10f, a3dd760ce6480a097, acc2ec8f1a893475c, a82ae365d19c988f4.
== STEP 1b · THE VM == `zsh tools/vm/vmctl.sh status` (perl alarm 60) in the game repo.
== STEP 2 · REVIEW + TRACK == TRACKER + make_tracker.py; LEDGER diff vs $CLAUDE_JOB_DIR/tmp/ledger_now.txt + clash checks with ctx.py + 58k_TRAPS.md; SHEETS reviewed myself + regrade queue; GAME DROPS merge flow (merge-tree, merge --no-ff, csc_check + build_test in background, PLAN + CHECKLIST, tf_merges + make_build_board, push v4 to farisbasha after green).
== SAFETY == never print/commit keys (.env, keys/, inbox token); kill only our own PIDs, never pkill/killall; push ONLY to farisbasha; no destructive git; `git add <paths>` only; plain commit messages; no store builds/installs; no sudo; headless Chrome only (--headless=new, private profile, sentry); one Unity at a time; never wrap build_test.sh in locked.sh; never merge into v4 while build_test runs; never prune alpha.eatiko.com; never cswap auto; no globbed/relative rm -rf, delete via python with absolute paths.
== STEP 3 · HUNG WORK == == STEP 4 · MEMORY == End with a 1-3 line status in chat AND as the inbox status line.