Axiom

Task

Execute the requested action: $ARGUMENTS

Calling rule: /axiom <action> <arguments>.

If the requested action is ambiguous or contradictory, do NOT guess: flag it and ask for clarification before executing.

This skill owns axiom-config.md — all other skills read it, only this skill writes it. Templates used below ship in this skill folder: claude-kernel-template.md, soul-template.md, brand-config-template.md, axiom-config-template.md, getting-started-template.md.

If action is empty

Describe this skill: its purpose and the actions table from "overview". Do not execute anything.

If action is "overview" (or unknown)

If the action was unknown, say so first, then show the table.

If action is "start"

A. Detect state

Read the workspace root. Three cases:

• Fresh (no axiom-config.md, empty folder or unfilled template copy): full flow below.
• Partial (axiom-config.md exists but scaffolding or module files incomplete): report what exists, run only the missing parts.
• Configured: report status and offer activate / tier / language instead. Never re-scaffold over a working system.

B. Intro

Deliver this positioning (adapt wording to the conversation language, keep the order):

1. Structure — "AxiomCore keeps everything you work on in one enforced system: numbered folders, consistent names, unique task IDs, and a wiki memory that mirrors it all. Nothing gets lost."
2. Two work trees — "Everything you do lives in one of two trees: projects are temporary — a bounded goal, worked toward and finished once (a website launch, a market analysis); workstations are permanent — recurring routines repeated on a cadence (monthly reporting, ads monitoring). A finished project can become a workstation when its work turns routine."
3. Disciplined workflow — "I never rush ahead: every task runs plan → your approval → step-by-step execution with validation. You stay in control."
4. Health — "/lint audits the whole system on demand: broken links, orphan pages, naming violations."
5. Growth — "/setup lets the system learn new capabilities without breaking the structure."

Then show the setup roadmap, marking what can be skipped:

Setup takes ~5 minutes: 1. Language (required) → 2. Simple or advanced mode (required, default simple) → 3. Naming convention (explained, customizable) → 4. Modules — Voice, Brand, Todos, Dev Planning (all optional — skip any, activate later with one command) → 5. Your current work → first project (optional). Skip anything optional; I'll suggest it later when it would help.

Every skippable question accepts "skip" and moves on without discussion.

C. Interview (one question at a time)

1. Language. Ask which language to work in; reply in the user's language from their first answer onward. Note in one line: folder and file names always use Latin-alphabet kebab-case regardless of language (any language written in Latin letters is fine; other scripts are transliterated).
2. Tier. "Keep it simple, or full control?" — simple: 2-level projects, numbering and IDs managed silently; advanced: full protocol, 3-level hierarchies, /dev. Default simple; switchable anytime with /axiom tier.
3. Naming convention. Explain the default convention briefly, scaled to the chosen tier — always: folder and file names are lowercase Latin kebab-case (market-analysis.md); every project and workstation folder carries an uppercase acronym prefix (ED-elena-design) so it is identifiable at a glance. Advanced tier additionally: sub-units and work packages / work areas carry two-digit number prefixes (01-marketing/02-seo-optimization), and those numbers form the task IDs (ED-01-02-05). Then one line on produced files, again scaled to the tier — simple: "documents I produce get a stable name with a running number, which I manage for you — no dates or version numbers in filenames, so links to them keep working"; advanced: the full pattern <acr>-<address>-<NNN>-<name>.<ext> (ed-02-002-keyword-analysis.md), the number from the work package's own counter, with version and date living inside the document. Close with one line: this convention is the system default — if the user prefers a different scheme, it can be customized later via /setup new (a setup file adapts the convention system-wide and consistently). Not a blocking question; continue unless the user objects.
4. Modules (multi-select, one-line pitches): Voice (drafts sound like you), Brand (colors/logo/tone applied to documents), Todos (task tracking, optional Notion board), Dev Planning (spec → plan → features; advanced tier only — selecting it from simple triggers the tier question again).
5. Current work. "What are you working on right now?" — free text, feeds step E.

D. Scaffold

Create in this order (skip whatever already exists):

1. Folders: projects/, workstations/, vault/me/, vault/projects/, vault/workstations/, vault/knowledge/, vault/sources/, setup/done/, .claude/templates/; plus brand/config/, brand/images/, brand/templates/ if M2-brand.
2. CLAUDE.md — render from claude-kernel-template.md: fill {{PLACEHOLDERS}}, include only the active modules' blocks and the chosen tier's blocks, strip all marker comments.
3. axiom-config.md from its template (language, tier, module flags, empty nudge log).
4. vault/index.md, vault/log.md (first entry: setup), vault/me/profile.md seeded from interview Q5.
5. GETTING-STARTED.md from its template.
6. .claude/templates/output-document-template.md — copy output-document-template.md verbatim (no placeholders to fill; its <...> markers belong to the documents produced from it).
7. setup/README.md — the setup area's register: Acronym: **SU**, Next setup: **001**, and one line stating the file naming rule (su-<NNN>-<name>.md, allocated by /setup new at drafting, kept when the file is archived to done/).

Then run the mini-interview for each selected module (same flows as "activate" below). Log everything to vault/log.md.

E. First win

From Q5's answer, propose concretely: bounded goal → "shall we create this as your first project?"; recurring activity → workstation. Restate the project/workstation distinction in one line so the user understands why this type was proposed. On yes, run the normal /project new project or /ws new WS flow — this shows numbering, the vault mirror, and the plan → approve → execute loop in action. On skip, point to /project and /ws.

F. Wrap-up

Summary table (language, tier, active modules) + up to three "try next" suggestions based on the interview + one line each: /axiom status shows this anytime, /axiom activate adds skipped modules, the vault/ folder can be opened as an Obsidian vault for browsing (optional), and feedback/bug reports are welcome at https://github.com/protonium-labs/axiomcore-plugin/issues.

If action is "activate"

1. Parse the module: M2-voice, M2-brand, M3, M4 (accept natural names: voice, brand, todos, dev). Unknown → list valid modules. Already active → say so, stop.
2. Run the module's mini-interview:
   • M2-voice: how the user writes (3 adjectives), formality preference, do/don't phrases, optional sample text to imitate → render soul.md from soul-template.md.
   • M2-brand: logo file (into brand/images/), primary/accent colors (or offer to extract from the logo), optional template.pptx/template.xlsx (into brand/templates/) → render brand/config/brand-config.md. Any item skippable.
   • M3: check whether a Notion connector is available. Available → offer the Kanban mirror (mode notion) or vault-only; unavailable → mode vault-only, one line that connecting Notion later upgrades it.
   • M4: requires advanced tier — from simple, run the tier dialog first (see "tier"). No interview; confirm the order spec → plan → feature and that this skill plans, never codes.
3. Insert the module's block into CLAUDE.md (from the kernel template, markers stripped).
4. Flip the flag in axiom-config.md, append to vault/log.md, confirm with one concrete usage example.

If action is "deactivate"

1. Parse the module; M1 cannot be deactivated. Not active → say so.
2. Remove the module's block from CLAUDE.md, flip the flag. Never delete user files (soul.md, brand/, tasks.md contents stay untouched) — say so explicitly.
3. Log and confirm.

If action is "tier"

1. Parse target tier. Same as current → say so.
2. simple → advanced: explain what unlocks (3-level hierarchies, exposed ID protocol, /dev availability, structural migrations via /setup). Swap the tier blocks in CLAUDE.md. Existing projects keep their depth — nothing is restructured.
3. advanced → simple: warn what gets hidden (not removed): 3-level creation, /dev. Existing 3-level projects continue to work; only new creation is limited. Confirm before applying.
4. Update axiom-config.md, log, confirm.

If action is "language"

1. Parse the language. Update Language: in axiom-config.md and the working-language line in CLAUDE.md.
2. Confirm in the new language. Remind in one line: existing content stays as written; structure names remain Latin kebab-case.

If action is "status"

Read axiom-config.md and report: language, tier, module table (active/inactive, M3 mode), nudge setting, last 5 nudge-log entries. One screen, no lecture. If modules are inactive, close with a one-line reminder that /axiom activate <module> turns them on.

After any action

• Append to vault/log.md (exception: status is read-only, not logged).
• Rendered/changed CLAUDE.md must be lint-clean and carry an updated Last updated date.

Notes

• Nudges (kernel "Recommendations" section) always route here: an accepted nudge runs activate; a declined one is logged in the config's Nudge Log with the date (30-day cooldown).
• The template repo is a convenience, not a requirement — start scaffolds a complete workspace in an empty folder.