Show HN: AxiomCore – a Claude plugin that keeps your projects structured

I'll pull the GitHub README and HN thread so the paragraphs stay grounded in real product details, not invented numbers.AxiomCore landed on Hacker News as a Show HN from protonium-labs: a Claude plugin (Claude Code and Claude Cowork) that turns the assistant into a project-and-routine management agent instead of a chat that answers and forgets. The public source is the GitHub repo at https://github.com/protonium-labs/axiomcore-plugin; the thread sat at 1 point and 0 comments when this signal was captured. Positioning is explicit: enforced structure for project managers, consultants, and anyone running projects or routines, not another free-form coding helper.

Mechanically, the system is workspace-local. State lives in the folder (`axiom-config.md`, `CLAUDE.md`, `vault/`) with no global plugin state, so a new folder is a clean install. Structure is fixed: numbered project folders (for example `01-launch`, `02-marketing`) and globally unique task IDs (for example `PW-01-03`). The agent runs a plan → your approval → step-by-step execution loop; structural changes do not land without approval. Memory is a plain-markdown, cross-linked wiki vault meant to be browsable in Obsidian. Skills cover onboarding (`/axiom`), projects (`/project`), workstations/routines (`/ws`), document ingest (`/ingest`), health checks (`/lint`), capability install (`/setup`), optional todos (including a Notion Kanban mirror), and optional Dev Planning (`/dev`: product spec → tech spec → delivery plan → feature documents). Optional modules also cover Voice (`soul.md`) and Brand. Two tiers exist: simple (2-level projects) and advanced (3-level hierarchies, exposed ID protocol, `/dev`). Subagents include a planner, reviewer, and vault-gardener. Install path is the Protonium marketplace plugin, then `/axiom start` for a short onboarding interview and scaffold. License is MIT; dev-planning methodology is adapted from the MIT-licensed Superpowers project.

For engineers and builders, the interesting claim is discipline as product: naming, placement, and registries are applied on every action, with `/lint` auditing broken links, orphan pages, naming violations, and registry drift. That targets a real failure mode of agentic coding sessions—context sprawl, invented paths, and tasks that appear without an explicit OK. The optional Dev Planning path is the bridge from “keep my PM workspace tidy” into software delivery artifacts that stay immutable once written. Because the vault is owned markdown rather than a closed vendor store, the structure survives session resets and tool switches.

Market context is the growing Claude plugin and skill ecosystem: marketplace install, skill commands, and subagents map onto how teams already extend Claude Code/Cowork. AxiomCore sits closer to personal operating systems and PM tooling (wiki memory, routines, approval gates) than pure code-generation plugins, while still shipping a software-planning module. Overlap with Superpowers-style planning workflows is acknowledged in the license notes; differentiation is the enforced hierarchy, ID protocol, and lintable vault rather than one-shot prompt packs. Early HN traction is thin (single point, no comments yet), so the signal is discovery-stage rather than consensus.

Practical takeaway: if your Claude sessions already create files and plans but leave folders inconsistent, evaluate AxiomCore in a throwaway workspace via the Protonium marketplace install and `/axiom start`, then run `/lint` after a real project scaffold. Watch whether the approval gate slows useful work or cuts silent drift; whether simple vs advanced tier matches how you actually name work; and whether optional modules (Todos/Notion, Brand, Dev Planning) stay optional in practice or become required for value. Next checks are GitHub Issues/Discussions activity, how well multi-workspace isolation holds when sessions leak memory, and whether the structure engine stays usable once a vault grows large.