Since this hype is just an overlay between 2 games, I built a real mod.
I put Jak from Jak 3 into Risk of Rain 2. It's an actual mod, and runs through R2modman. You can see his gun, slider, and even dark Jak.
will be polishing this more as I go
now the verbose version:
- switch providers and accounts mid-thread with context handoff
- switch back to previously used providers
- change models and options between turns
- remember options per provider and model
- switch providers in existing mobile threads
- queue provider switches without stopping current work
- apply queued switches in delivery order
- recover failed native resumes with portable context
- show provider/model changes in the timeline
- transfer short conversations intact
- select relevant history from longer conversations
- preserve message order and attribution during handoffs
- carry partial failed/interrupted work into handoffs
- keep new prompts separate and unshortened
- let agents retrieve omitted history
- inject native history into supported codex versions
- reject handoffs that cannot fit available context
- configure handoff history budgets
- exclude private reasoning, live tool state and historical attachments from handoffs
- fork from a selected finished run
- choose a fork's provider on its first message
- delay fork session creation until needed
- use native forks where supported
- use portable context for cross-provider forks
- fork codex at the selected turn
- fork claude from earlier turns
- fork pi into another workspace
- fork cursor through portable context
- preserve inherited history in forks
- keep claude forks usable after source rollback
- fork failed, interrupted, cancelled and limited runs
- fork before checkpoint finalization when provider work is finished
- reject active or already-rolled-back fork sources
- merge fork context back into its source
- expose fork and merge-back tools to agents
- wait for mobile fork state before navigation
- delegate tasks across providers and models
- track delegated tasks as child threads
- choose child provider, model, options and role
- inherit compatible parent settings
- give children explicit task context rather than copying parent history
- delegate asynchronously or wait
- include nested work in task completion
- keep children running after wait timeouts
- wake parents when children finish
- batch eligible child results
- deliver later child completions in subsequent wakes
- preserve original results across follow-up work
- inspect, follow up with and cancel delegated tasks
- distinguish working, waiting-for-children and result-ready states
- reject unavailable delegation targets
- prevent child permission escalation
- discover live provider/model capabilities through agent tools
- create ordinary threads individually or in batches
- create idle threads or launch them with prompts
- launch threads in project roots or worktrees
- choose worktree base branches for stacked work
- launch scratch threads through agent tools
- include attachments in thread launches
- list and filter project threads
- search thread titles and content
- read messages, plans and activity across threads
- page through history and long items
- send, queue, steer or restart another thread
- wait for a selected run
- request thread interruption
- rename threads and regenerate titles
- link and unlink pull requests
- inspect lineage, workspace, run and request state
- attribute agent messages to their source threads
- allow cross-project reads through human-attached references
- keep ordinary thread access project-scoped
- expose retry keys on supported agent operations
- distinguish interruption requests from confirmed stops
- let agents inspect, edit, cancel and reorder queues
- let agents promote queued messages to steering
- let agents answer supported questions, not permission approvals
- let agents change their own provider/model/options
- let agents pin, snooze, settle, archive and mark threads unread
- let agents create and manage scheduled tasks
- let agents manage projects and clone repositories
- expose name-only project creation to agents
- let agents prepare, send and discard pending attachments
- let agents inspect and hand off into worktrees
- rebind threads and restart sessions during worktree handoff
- expose scoped environment preferences to agents
- let agents list and close previews
- add first-class pi integration
- support pi 1.0
- reuse pi installations and authentication
- configure pi executable, environment and arguments
- discover pi models and thinking levels
- preserve pi's default thinking level
- support pi native resume, fork, rollback and steering
- integrate pi with t3's queue and context handoffs
- show pi context usage
- expose pi skills in the picker
- load pi user/project extensions and context
- render pi extension dialogs in the composer
- show pi extension notifications in the work log
- bridge t3 tools into pi
- show pi example-subagent progress and results
- fall back to pi default when discovery stalls
- enforce pi tool approvals by permission mode
- hide unsupported pi auto mode
- resume pi sessions after permission changes
- detect opencode 1.x versus 2.x
- add opencode 2 runtime support
- retain limited opencode 1.x support
- launch local or connect to existing opencode 2 servers
- render opencode 2 text, reasoning and tools
- track opencode 2 models, history, context and usage
- handle opencode 2 approvals and structured questions
- map permission modes to opencode 2 rules
- support opencode 2 native steering, forks and rollback
- support opencode 2 compaction and plan mode
- discover opencode workspace/session inventory
- show opencode child sessions as subagent threads
- route child requests to parents
- preserve opencode background children and parent wake-ups
- inject t3 tools into local opencode 2 servers
- reconcile opencode sessions after restart
- reject unsupported opencode question fields explicitly
- fix opencode prompt-admission and cancellation races
- fail broken opencode streams instead of hanging
- clean up orphaned opencode servers
- move cursor execution to its official sdk
- add cursor browser login across clients
- store cursor credentials per execution-environment instance
- discover cursor sdk models
- support cursor model changes between turns
- load cursor project skills and rules
- support cursor images, reasoning, plans, todos and tools
- inject t3 tools into cursor
- resume cursor local agents
- implement cursor steering through interrupt/restart
- display cursor native children as read-only threads
- enable cursor sdk retries
- recover abandoned cursor runs and shell-start failures
- replace cursor cli/endpoint controls with sdk configuration
- use cursor sdk sandboxing in restricted modes
- add acp registry search and onboarding
- install compatible registry agents on the selected server
- verify registry checksums when supplied
- support existing executable overrides
- share browser/terminal authentication across clients
- support mobile terminal-auth responses
- retry, cancel and verify provider authentication
- discover acp models, options and native commands
- map advertised acp plan/build modes
- refresh pickers after runtime configuration changes
- negotiate newer acp with older-version fallback
- render acp usage, reasoning, plans and compaction
- list/import native sessions where supported
- deduplicate repeated native-session imports
- preserve user-renamed titles during imports
- render acp resources and explicit unsupported-media placeholders
- show old/new-text file diffs
- distinguish file search from web search
- expose t3 orchestration to compatible acp agents
- improve devin terminals, questions and child-agent presentation
- use allow-once for automatic acp approvals
- use fresh sessions where native acp rewind is unavailable
- preserve claude continuity through resume, wake and idle release
- preserve claude steering state and model options
- allow claude steering despite next-turn option changes
- restore claude questions, plans, todos and implement actions in v2
- support claude questions in plan mode
- restore claude resume compaction
- correct claude post-compaction usage
- preserve claude read-image previews
- separate claude parent and child usage
- keep child output and reasoning in child threads
- preserve resumed child identity and requested models
- correct nested claude child ownership and wake routing
- stop completed claude children appearing active
- separate claude background replies from queued prompts
- prevent background turns prematurely completing queued compaction
- label claude monitors correctly
- preserve claude limit-reset information
- remove the old short timeout on long claude t3-tool calls
- expand home-directory shortcuts in provider executable paths
- fix codex routing through shared app-server sessions
- unload detached codex threads
- support codex rollback after app-server restart
- correct codex native and nested subagent mapping
- route codex child approvals to parents
- track codex background completions and resumed children
- fix codex startup-stop races and stuck waiting states
- preserve codex reasoning and tool timestamps
- restore codex todos and actionable plans in v2
- honor codex launch arguments
- port codex feedback and managed authentication to v2
- fix codex interrupted sign-out recovery
- remove unsupported grok auto-accept-edits mode
- ask about grok classifier-rejected commands
- correct grok approval scopes
- show grok api failures instead of empty success
- track grok background replies as continuation runs
- share antigravity account authentication across clients
- send antigravity images directly and other attachments as file paths
- replace browser-owned queues with durable server queues
- preserve queue order across restarts
- hold restarted queues for explicit resume
- preserve queues through usage limits
- edit, reorder, remove and steer queued messages
- show queue previews and attachment thumbnails
- retain generic files during queue edits
- add/remove attachments while editing
- keep queue edits pending until saved
- restore previous composer drafts after queue editing
- handle queue entries changed by another client
- stop held queues falsely showing working
- hide internal wake deliveries from user queues
- run limit recovery before queued follow-ups
- recover late steering as follow-up work
- add shortcuts for alternate send, queued steering and queue editing
- add a background-thread/new-composer shortcut
- make new shortcuts remappable
- add mobile queue editing and reordering
- expose mobile queue/steer preferences and alternate send
- track whether approvals/questions remain answerable
- disable dead-process approval actions
- distinguish blocking and asynchronous questions
- allow asynchronous answers after provider exit
- retain answers on question timeline items
- remove answered questions from pending controls
- preserve select/multiselect semantics across clients
- hide disallowed custom-answer inputs
- route native-child requests to parents
- prioritize pending approvals over working status
- keep mobile stop controls accessible above questions
- expire dead callbacks without discarding all asynchronous questions
- rebuild timelines around stable server-owned items
- group multiple provider turns into one logical response
- preserve streamed-item order
- page older history with loading/error states
- deduplicate paged and live history
- avoid resurrecting hidden history items
- improve cold-open and queued-run scrolling
- group work, reasoning and subagents into collapsible sections
- show handoff, fork, compaction and interruption events
- label queued and steered messages
- link agent-originated messages to their source
- summarize meaningful tool actions
- avoid counting retries/failures as completed work
- keep stopped commands and failed runs visible
- identify background completions by task type
- anchor timers to the correct run
- freeze settled subagent durations
- remove legacy token-by-token streaming settings
- show command metadata and diff navigation rather than raw tool-output bodies
- make lineage durable and navigable
- show child models, status, progress, results and duration
- stabilize lineage ordering and parent status
- group current/previous children and retain failures
- page large lineage lists
- add collapsible multi-subagent cards
- add a mobile agents sheet
- replace native-child composers with status and open-parent controls
- filter native children from ordinary sidebar/search navigation
- suppress redundant child completion notifications
- replace the separate agents panel with lineage/inline views
- move workspace/git/script controls into thread details
- add workspace, version-control, automation and lineage sections
- show saved project scripts and linked pull requests
- keep merge controls aligned with current checks
- adapt thread details between inline/popover layouts
- coordinate chat, details and preview sizing
- keep previews from covering the composer where possible
- expose composer workspace controls with a persistent-display preference
- add thread references through title search and sidebar drag/drop
- make reference chips open source threads
- retrieve referenced history on demand
- separate context occupancy from per-turn usage
- preserve usage through incomplete terminal events
- correct usage after compaction and session changes
- add a distinct limited status
- add manual resume and resume-at-reset
- add snooze-until-reset separately from automatic resume
- configure recovery defaults across clients
- recover overdue continuations after restart
- invalidate obsolete continuations after new work/archive/settlement
- avoid inventing reset schedules when timing is unknown
- make stop work during startup
- let ui stop reach background work after foreground completion
- avoid useless stop controls in queue-only states
- reconcile stale runs and children after restart
- tell agents which background work died
- preserve model/options in eligible restart continuations
- recover eligible settled threads that lost background work
- attribute background replies to their own continuation runs
- let persistent dev servers coexist with completion alerts
- bind checkpoints and diffs to v2 runs
- correct repeated rollback boundaries
- include failed/interrupted/cancelled runs in rewind handling
- fix edit-from-here after stopped runs
- surface rollback failures instead of hanging
- reject unsafe checkpoint restores
- preserve conversation-only rewind
- check worktree overlap, nested paths and symlinks before file restore
- include archived threads in workspace-safety checks
- avoid confusing shared provider directories with thread workspaces
- preserve task-step timing across restarts
- add scheduled-task settings across web/mobile
- create, edit, pause, resume, run and delete automations
- filter automations by project/environment
- use execution-environment timezones for fixed schedules
- validate mobile schedules and warn about unsaved edits
- configure static, generated or custom worktree branch names
- allow project branch-naming overrides
- retain temporary branches when generated names fail
- derive sidebar status from actual runtime state
- distinguish working, waiting, limited and failed
- stop waiting timers growing like active-work timers
- preserve failures despite pending background work
- avoid hiding active work behind newer queued/cancelled runs
- synchronize visit/read state where supported
- propagate metadata changes across clients
- prevent duplicate title regeneration
- make settlement server-owned
- account for linked prs, pins and background work in settlement
- use actual activity for mobile archive eligibility
- reduce sidebar churn from hidden child output
- bound snapshots and page long conversations
- reduce transcript decoding and retained memory
- coalesce streaming updates and narrow subscriptions
- reduce unrelated rerenders and settled-agent timer work
- improve mobile scrolling, anchoring and overlay bounds
- persist v2 client caches
- tolerate unknown compatible event types
- migrate once into a separate v2 database
- keep settings, attachments and workspace files shared
- use a separate desktop browser profile
- import old messages and metadata without promising full runtime-history migration
- resume imported threads through fresh sessions and portable context
- load large imported transcripts progressively
- reject incompatible client/server versions with update guidance
- require compatible native mobile builds for native changes
With everyone combining different games and mixing/matching them i set out to try something a bit more niche.
Whats the FPS ceiling on minecraft for mac?
The answer? ~1024 median fps at full native 5k res on my Apple Studio Display.
I compared this to 20 of the top Minecraft performance mods and my singular perf mod is 8x faster at 0 quality cuts.
Default settings.
Itβs pretty insane to think that I have @NousResearch Hermes desktop app and Pi running no problem on my 2010 MBP via @dhh Omarchy - but I canβt even get chrome or safari to run properly on MacOS because Apple only supports High Sierra for those older laptops. If you have old gear sitting around that Apple and friends want you to get rid of so you buy more; Omarchy is your friends
It's 4am. I just finished a project I've wanted for awhile.
This is Slopalytics. It's a much better view for Artificial Analysis data. It also has a bunch of usage data from T3 Code.
I plan to add much more in the near future, including some new benchmarks :)