@nicksdot@brendt_gd Yes, ecosystem, as well as opinionated frameworks which put PHP ahead of the agentic curve, and moderately priced senior engineering talent for the parts which require human intellect and experience.
I just released Expressive!
It provides typed objects for Eloquent, converting Eloquent models into typed PHP objects and converting those objects back into Eloquent models when the application needs to persist data.
Eloquent remains responsible for querying, relationships, casts, visibility rules, mass assignment, and database writes. Expressive gives the rest of your application a typed object boundary without replacing Eloquent.
Make your Eloquent models Expressive! 🔥🚀
Repo and docs links in the thread! 👇
It's amazing. Provides a communication and control layer between cli agent instances -- interagent communication, shared to-dos, scratchpads, key:value stores, process list, etc. -- as well as the application itself.
Opens up _all_ the possibilities while being unopinionated -- pure affordance.
@nicksdot The meta-harness provides affordances, but is not prescriptive -- it's just a communication layer between agent instances. The skill (and supporting hooks, functions, etc) is where the orchestration is codified.
Solo exposes orchestration primitives via mcp, api, and cli -- so I am able to have a workspace orchestrator agent spin up a Solo project, add a task orchestrator, feed it a prompt as if I typed it into the cli (valid use of Anthropic subscription), have that agent spin up agents which live in their own instances (less limiting than native subagents), communicate with them bidirectionally, and those agents can create shared to-dos and scratchpads, share a key:value store, set timers, act on the idle notifications of others, etc.
Together with the orchestration skill that I'm iterating on and my own cli which handles setting up workspaces, worktrees, loading agentfiles and skills based upon a manifest that translates issue labels to install bundles, running various setup steps, and spinning up Solo terminals within the project for the various long running processes an agent might want during a task, there's not much that I can't do.
Absolutely -- "making deterministic what can be deterministic" is spot on. I've been calling all of that "the scaffold". Models are plenty capable and harnesses are converging (and Solo is the meta-harness that I didn't know I needed) -- I think that's where the biggest gains are right now.
@nicksdot Except for the the fact the the skill is lazy about colons, it seems. Gonna pain me to have to add a function for something as simple as creating agent names, but...
My keyboard is covered in dust.
This was a series of steps. Initial draft was Claude and Superpowers (brainstorm and plan-writing) followed by a Codex contrarian review. Dogfooded and had a few skill polishing rounds. Then I went back to Superpowers and we added logging and created an eval — a simple state machine and a skill which runs the logs, looks for process violations, and grabs a segment of the session history surrounding the violation. Finally, another skill which compares those to the “SoloTerm Orchestration” skill and suggests hardening steps. Lather, rinse, repeat.