Per-feature build plans for atom. One file per feature. Each captures locked decisions, rationale, and what to watch for at build time.
These are the source of truth for the upcoming build. Read the relevant file before starting work on that feature.
atom-setup.md. Interactive wizard. The first thing a user runs after cloning atom. Drives all per-project configuration.nucleus.md. Cross-project learning store. Lives at~/.atom/nucleus. Captures session learnings; syncs to private GitHub.learnings.md. User-owned playbook at~/.atom/learnings/. Where generalized nucleus entries land and ride forward into every bootstrapped project.docker.md. Containerization tiers. Fully optional; smart default based on stack choice.model-race.md. Parallel model comparison workflow using Git Worktrees. Niche power-user feature; opt-in.
v0.2.md. Milestone plan. 10 items across 3 waves; sequenced so the cheapest wins land first and the architectural shift lands last. v0.1.2 shipped the install-bug patch ahead of this milestone.distribution.md. How atom is installed, upgraded, and separated from project clones. Locks the architectural decisions behind v0.2 Wave 3 (~/.atom/atom/install,atom upgrade,install.shcurl one-liner, auto-update check).stack-presets.md. Five new stack presets — Python/FastAPI, Swift (Vapor), Rust (Axum), Go CLI (Cobra), TypeScript library — closing the v0.1 gap where only Next.js had a real preset. v0.2 Wave 2.
v0.3.md. Candidate to-do list, not locked. Bundles 7 new candidates (hook-driven nucleus capture, prompt auto-injection, PreCompact/PostCompact,atom doctor, SQLite+FTS5 nucleus storage, research/wiki layer, cost tracking) sourced from therohitg00/pro-workflowcomparison analysis, plus the two pre-existing v0.3 deferrals below. v0.3 planning session decides yes/no/defer on each.model-race-auto.md. Pre-existing deferral. The--autoflag formodel-race start. Cross-CLI parallel session orchestration via a per-CLI adapter contract. Risky (per-CLI maintenance surface); parked to v0.3 with the design locked.nucleus --semantic. Pre-existing deferral. Local embeddings via@xenova/transformers. Risky (embedding-model + vector-store + indexing decisions). If v0.3 picks up SQLite migration (#16), this rides on top.
v0.1 (locked, built): nucleus → learnings → docker → model-race.
atom-setup built incrementally as features came online.
v0.1.2 (shipped 2026-05-07): fresh-clone install crash fix.
v0.2 (locked 2026-05-07, in build):
Wave 1 — foundation, parallelizable, ~1 day
#2 VERSION + atom upgrade
#3 ./atom-setup --reinstall flag
#4 nucleus schema migration tooling
#5 live gh repo create
#8 Migration doc for in-place users
Wave 2 — user-facing features, parallelizable, ~3 days
#6 Stack presets (5 new)
#7 speckit-constitution integration
Wave 3 — distribution architecture, sequential, ~2 days
#9 ~/.atom/atom/ install + atom-setup new <name>
#10 install.sh curl one-liner + README rewrite
#11 bin/atom-update-check + snooze
Wave 3 is preauthorized to de-scope to v0.2.1 if it hits a design wall (most likely the in-place migration story).
v0.3 (candidate list, not locked):
See v0.3.md for the full list. Nine items total:
- #12 Hook-driven nucleus capture (
Stophook) - #13
UserPromptSubmitauto-injection of nucleus + learnings - #14
PreCompact/PostCompacthooks for nucleus - #15
atom doctorcommand - #16 Move nucleus to SQLite + FTS5
- #17 Research / wiki layer
- #18 Cost tracking surface
- #19
model-race --auto(pre-existing deferral, design locked) - #20
nucleus --semantic(pre-existing deferral, design locked)
The v0.3 planning session decides yes/no/defer on each candidate.
#12–#18 came from the rohitg00/pro-workflow comparison analysis;
#19–#20 carried over from v0.2 with their design already locked.
All v0.1 plans built. All v0.2 plans locked (2026-05-07). v0.3 items have full design locked but build is deferred. Decisions reached through guided planning. No re-litigation without a written reason.