I have a "docs" folder in every project. Every iteration on the project generates a plan (if I'm using superpowers, it generates two, the design doc and the implementation plan). Those go into docs.
Inside the docs folder is an archive folder. Stale and completed plans and docs get moved there.
I generate a schema.md that describes the database schema in LLM-friendly terms that lives in docs.
If there's a gnarly design or implementation design made that the LLM keeps tripping up on, I get it to document that and put it in docs.
The handoff.md doc lives in docs, and gets moved to archive once it's no longer relevant. Same for todo.md
My agents.md instructs the LLM to read the docs but ignore the archive, and acknowledge the docs it has read. So the start of every session gets a "I have read <date>design.md, <date>plan.md, schema.md, notes.md, handoff.md". It's a decent chunk of context but as the OP says, this is necessary.
I also have a skill called "assessing project drift" that looks at the archive and assesses the project on how far the code has moved from the intention, if there are any chunks of code that were relevant but have been superceded and can be removed, and if there's any egregious tech debt that should be dealt with now. The docs archive is really useful for this.
Okay, so using superpowers. Have you actually looked at the superpowers? They're garbage. That shit was maybe helpful a year ago. It's now just cluttering up space.
I think a lot of people are stuck on patterns that "used to work" (and I use that loosely because I really feel like LLMs are getting work done IN SPITE of using all these additional skills/plugins/etc). Go ahead, delete all that shit and try building without. Yeah, sometimes you may have to hand hold the LLM, but that hand holding is a LOT less than doing whatever this is.
This all just sounds like a rube goldberg machine you've made.
I took the original version of superpowers and have iterated on it over the last year or so (actually only around 9 months but it feels longer!), so it's now more a thing of my own creation than the original (and definitely not what it has morphed into, which I agree is a pile of crap now).
marcus_holmes · · focus · HN ↗
Inside the docs folder is an archive folder. Stale and completed plans and docs get moved there.
I generate a schema.md that describes the database schema in LLM-friendly terms that lives in docs.
If there's a gnarly design or implementation design made that the LLM keeps tripping up on, I get it to document that and put it in docs.
The handoff.md doc lives in docs, and gets moved to archive once it's no longer relevant. Same for todo.md
My agents.md instructs the LLM to read the docs but ignore the archive, and acknowledge the docs it has read. So the start of every session gets a "I have read <date>design.md, <date>plan.md, schema.md, notes.md, handoff.md". It's a decent chunk of context but as the OP says, this is necessary.
I also have a skill called "assessing project drift" that looks at the archive and assesses the project on how far the code has moved from the intention, if there are any chunks of code that were relevant but have been superceded and can be removed, and if there's any egregious tech debt that should be dealt with now. The docs archive is really useful for this.
kaydub · · focus · HN ↗
I think a lot of people are stuck on patterns that "used to work" (and I use that loosely because I really feel like LLMs are getting work done IN SPITE of using all these additional skills/plugins/etc). Go ahead, delete all that shit and try building without. Yeah, sometimes you may have to hand hold the LLM, but that hand holding is a LOT less than doing whatever this is.
This all just sounds like a rube goldberg machine you've made.
marcus_holmes · · focus · HN ↗
And yes, very much a Rube Goldberg machine ;)