memory-init

Memory Init

Interactive onboarding. Mechanism only — option lists come from templates/<archetype>.yaml. The skill asks one question per turn, collects answers, then writes five frontmatter-keyed memory files plus the CHANGELOG seed.

Guard

Read the state of memory/ first.

• Missing or only README.md / CHANGELOG.md → proceed.
• Any of profile.md, targets.md, platforms.md, patterns.md, on-horizon.md already has non-trivial content → halt and confirm with AskUserQuestion. Options: overwrite, extend, abort. Default recommendation: abort — route to memory-curator for incremental updates.

Tool rules

• One question per turn via AskUserQuestion. 2–4 options each. The tool auto-provides "Other" (free text).
• Numeric answers: surface 3 plausible ranges from the template; "Other" lets the user enter exact values.
• Never invent. Skipped fields land as null in frontmatter so skills can detect missing config.
• Question wording: terse and structural. Translate option labels to the chosen output_language after Block D.

Step 0 — Pick archetype

Single AskUserQuestion. Options are the three archetype files in templates/:

• equity-heavy — index/ETF investor; equities-first; modest crypto
• crypto-native — crypto-first; BTC/ETH core; stablecoin yield
• mixed — balanced multi-asset; ETFs + stocks + crypto + yield

Read the chosen templates/<archetype>.yaml. All subsequent option lists come from this file.

Steps 1-5 — Blocks A–E

Iterate the field tables in references/block-fields.md, one block per step, one AskUserQuestion per field. The reference covers Block A (profile.md), Block B (targets.md), Block C (platforms.md), Block D (patterns.md), Block E (on-horizon.md) — each row maps a frontmatter field to its option source under templates/<archetype>.yaml.

For Block D's clarification_triggers, offer the catalog in references/clarification-triggers.md as a multiSelect.

After Block D, switch all subsequent option labels and confirmation prompts to the chosen output_language.

Step 6 — Review and confirm

1. Show captured answers compactly: one line per top-level field group.
2. AskUserQuestion with options: Write all files / Edit a section / Abort.
3. On "Edit a section" → re-ask only that section.
4. On "Write all files" → proceed.

Step 7 — Write

Render YAML frontmatter per the schema in memory/README.md, then a markdown body with section headings reflecting the captured fields. Write in this order:

1. memory/profile.md
2. memory/targets.md
3. memory/platforms.md
4. memory/patterns.md
5. memory/on-horizon.md
6. memory/README.md — only if missing
7. memory/CHANGELOG.md — seed entry: ## YYYY-MM-DD\n\n- **Initial scaffold:** memory/ created via memory-init using <archetype> archetype.

Step 8 — Closeout

Return to chat in the user's chosen output_language:

• Confirmation that memory/ is populated.
• Path list of files written.
• Suggested next step: run portfolio-advisor for the first report, or use memory-curator to refine.

Failure modes to avoid

• Don't skip the guard. Overwriting non-empty memory destroys curated context.
• Don't batch questions. One question per AskUserQuestion call.
• Don't fabricate. Skipped fields land as null.
• Don't translate user free text. Preserve exact wording.
• Don't commit to git. That is the user's call.
• Don't bake new archetype-specific values into this file. New archetypes get a new templates/<name>.yaml, not a SKILL.md edit.