Added subagents and workflows to my pi setup
They're far from perfect, gonna spend a bunch of time with them tmrw cleaning up the prompts then I'll actually show them off, but at least for a first pass the actual impl feels pretty solid
Harness v1.1.0 is out. the native macOS terminal built from the metal up. two big additions 👇
1/ Remote + headless daemon. Run the daemon on another box (Linux included) and drive it from your laptop over plain SSH: harness-cli new-session --host devbox.
No new keys, no new crypto. It reuses the SSH trust you already have.
2/ Persistent scrollback. Your history is written to disk per pane and restored when the daemon restarts, so a crash won't wipe it. Plus a 1.5–1.66× faster parse engine and a rebuilt settings window. (oh, and an agent notch to see what your agents are up to)
The official /codex plugin for Claude Code is pretty sweet.
/codex:review to have Codex critique CC's work
/codex:adversarial-review for increased spice
/codex:rescue to handoff a mess only GPT 5.5 can untangle
It's like /handoff with less ceremony. Good in a pinch!
People of Pi! I went absolutely mental. Massive update to pi install npm@@howaboua/pi-smart-btw
- supports 9 dynamically-allocated /btw ephemeral sessions in one UI
- every session can be followed up
- btws inherit skills & extensions
- Pi shows replies in chat, hidden from the Clanker
- some visual feedback in the widget
- sessions stay alive until you shut down main process
- replies saved in jsonl file
- supports "dumb resumption" after closing Pi aka pasting previous questions and answers (will always use latest slot history if you overwrite one)
/btw to start/follow-up current slot
/btw 1-9 to start/follow-up selected slot
alt+1-9 to switch currently displayed slot
alt+z to insert /btw into prompt field
alt+c to insert Q&A into main agent & clear slot (history persists, but cannot continue)
alt+x to clear slot (as above)
alt+j/k to fold/unfold
alt+h/l to scroll slots
I don't think any harness in the world has a better btw functionality xDDDDDD
Happy Clanking!
introducing pi-dynamic-workflows
This is probably going to be a bigger token burner than pi-goal, BUT, dynamic workflows is the first implementation of subagents that i don't hate, mainly because it's "code mode" for subagents. agent writes a js-based workflow DSL into a dedicated tool, engine parses the workflow code and runs it.
the dsl implements some primitives for the agent (agent(), parallel(), pipeline(), phase() and log()) to keep it as simple as possible.
now available in @badlogicgames pi!
pi install npm:pi-dynamic-workflows
I think you’ll really like Opus 4.8
It’s as smart as its benchmarks show but expresses and utilizes that intelligence in a warm and collaborative way.
Workflows are a great way to utilize it- I’m hooked. Article on that soon.
Me again! Update to pi-markdown-workflows because nested AGENTS.md appended to the system prompt can get nasty.
pi install npm:@howaboua/pi-markdown-workflows
What this extension does:
- /skills - browse, create, use, refine, or delete agent skills
- /workflows - browse, create, use, refine, append, promote to skills, or delete repo workflows
- /learn - capture useful session findings into the right AGENTS.md
- workflows_create - agent tool for writing reusable SOPs to ./.pi/workflows/<slug>/SKILL.md
- nested AGENTS.md autoloading for subdirectory-specific instructions
- bundled skill-creator skill certified by yours truly
The nested AGENTS.md loader watches successful file/path discovery tool results:
- read
- ls, find, grep
- shell-ish commands like bash, exec, exec_command, shell
- read/list/search-ish commands inside shell input: cat, sed, rg, tree, git, grep, git ls-files, etc.
For each touched path, it walks upward to collect the relevant nested AGENTS.md chain, skips the cwd root AGENTS.md, dedupes against active branch state, and appends the matching context to that specific tool result.
Appending it like that preserves caching better and prevents the Clanker from constantly referencing stale context if you jump repos.
https://t.co/wyFQIVfE0A
Happy clanking!
Two new skills this week!
- /handoff compacts your current session to a markdown file
- /prototype helps you prototype anything - UI or backend
And a bunch more updates. Here's the changelog:
I move fast. Next version with file tree thanks to @pierrecomputer
Trying to figure out a workflow to enable editor-like operations via patching in diff view without creating a separate editor view and turning the app into an IDE.
Soon™
https://t.co/7l8dtyNaqH
I have shamelessly stolen pi-diff-review from @badlogicgames and made it work for pi, @claudeai and @OpenAI codex. Introducing slop-review.
https://t.co/ZfTFTAR1Lu
Top 2
@FactoryAI’s @droid
- clean design
- best compaction
- very stable
- missions
- /wiki & /readiness-report & other skills
- model agnostic with optimisations for model families
@badlogicgames’s https://t.co/sxGYmdolH6
- super minimal
- strong SDK
- best to learn
- extremely stable
Since everyone's releasing a desktop app, here's mine. Built around @badlogicgames's Pi.
Heavily optimised towards coding with agents on YOLO. No file viewer, because it's not an IDE. No turn-by-turn diffs, because I do not believe anybody reads them.
Just you and your composer. And the Clanker.
Might be rough around the edges, but it's a diamond and will only get better. Built on Arch, btw. Works on my machine, but should also work on yours.
- Code view - with known suspects: composer, git-operations and terminal.
- Desktop chat mode optimised around UX/readability.
- Git-ops composer with diff review mode, comments-based review-flow & more.
- Pi-TUI mode that embeds Pi "as it is" inside of the app.
- Sidebar with projects/sessions management.
- Local dictation via sherpa-ONNX (CPU - it's fast enough)
- Browse, install, remove, and configure Pi skills/extensions.
Get it here https://t.co/xqpZHHm24W
Or npm/bun i -g howcode
You can just build things, apparently.
Unleash your Clankers. Vouch in place, lol.
Working on an open-source email client:
1️⃣ Hackable to make it your own.
2️⃣ Native Swift app (100mb RAM vs 2GB RAM for Superhuman).
3️⃣ View all inboxes in one place.
Still early, but give it a star here: https://t.co/QQKfTFVavt
Nearly 23K stars for a collection of markdown files I wrote
I guess they must be pretty good
I want to invest more time in this repo. So, folks who starred it, what can I do to make these skills more obvious to you?
- A docs site for the skills?
- Send them to plugin marketplaces?
Help me help you
https://t.co/64UuxC8V0T
Pi is beautiful. Told it to install agent-browser, and to build itself an extension.
Build your own personal assistant on top of pi. It’ll be infinitely more flexible than OpenClaw, Hermes could ever be.
Bringing the power of pi to 2 million Salesforce Developers. Going through paperwork/legal approval to opensource extensions under https://t.co/n4Wmzut7uX
@badlogicgames@mitsuhiko