PRDs: agentic build loop + installable coding-agent brain - #28
Draft
twilson63 wants to merge 1 commit into
Draft
Conversation
Two sequenced PRDs for evolving Build's agent into a developer agent: - docs/agent-loop-prd.md — turn the single-shot generator into a tool-using loop that runs the build, reads its own errors, and iterates until the app works. The competence unlock; build first. Grounded in the WebContainer hands that already exist (mountAndInstall, readFilesFromContainer, writeFileToContainer, startShell). - docs/installable-brain-prd.md — identity, soul, user instructions, and long-term memory via an installable brain: base skills -> per-user fork -> learned memory, hot in IndexedDB, durable/portable in ZenBin, content hot but signing authority server-side. Built on the Karpathy LLM-wiki pattern. Explicitly gated on the loop PRD. Note: reads the requested 'HyperDesk adapter' as hyper-zepto (the Scout Live data adapter); flagged in the PRD for confirmation. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Two sequenced PRDs for turning Build's agent into a real developer agent, from the design discussion. Draft — for review/comment, not merge.
Read order
docs/agent-loop-prd.md— the competence unlock. Turn the single-shot{reply, patches}generator into a tool-using loop that runs the build in the WebContainer, reads its own compile/preview errors, and iterates until the app actually works. Grounded in the hands that already exist (mountAndInstall,readFilesFromContainer,writeFileToContainer,startShell) — it's wrapping + orchestration, not new runtime infra.docs/installable-brain-prd.md— identity, soul, user instructions, long-term memory via an installable brain: base skills → per-user fork → learned memory; hot in IndexedDB (no HTTP on the hot path), durable + portable in ZenBin; content hot in the browser, signing key server-side. Built on the Karpathy LLM-wiki pattern. Explicitly gated on the loop PRD — a brilliant brain feeding a blind generator is wasted.The through-line
The original question was whether the agent should have identity/soul/memory. The PRDs answer it by inverting the order: competence first, soul last. The 'soul' isn't a persona — it's the installed, grown, portable brain itself, earned through a competent loop.
The brain PRD reads the requested 'HyperDesk adapter' as hyper-zepto — the data adapter Build's apps actually use with Scout Live. HyperDesk is a separate clawcloud remote-session client, unrelated to the WebContainer→publish pipeline (assumed voice-transcription). Flagged in the doc; correct me if a different system was meant.
🤖 Generated with Claude Code