Migrate

Assess an existing .minerva/knowledge/ corpus against the current LLM-wiki structure and report a migration checklist. minerva:migrate is read-only: it inventories what needs migrating and names the skill that closes each gap, then stops.

Read-only contract. This skill must not modify any file. Its allowed-tools omits Edit / Write / MultiEdit by design. It performs no renames, authors no cross-references, and runs no remediation skill — it only reports and recommends.

This is a SHAPE check, NOT a HEALTH check. migration_status tells you whether the corpus is in the shape the wiki tooling requires (conforming filenames, index + overview present, entries carrying ## Related). A clean migration inventory can still coexist with minerva:lint errors (e.g. a stale index watermark or catalog drift), so a passing migration check still requires a green minerva:lint pass and a minerva:synthesize pass — those are the ongoing health checks; this is the one-time shape audit.

Why this exists

The detector (scripts/knowledge_lint.py), the fixer (scripts/knowledge_fix.py), and scripts/synthesis_status.py all enumerate the corpus through the ENTRY_RE glob (^\d{3}-[a-z]+-.+\.md$) only. A file that doesn't match — a legacy entry named before the NNN-type-slug convention — is invisible to every one of them, so a pre-conventions corpus reads as a false clean across the whole toolchain. minerva:migrate is the one surface that globs the complement of ENTRY_RE and inventories those invisible files, turning a false-clean legacy corpus into an actionable checklist.

Target

The .minerva/knowledge/ corpus of the current working tree, resolved from git rev-parse --show-toplevel — the same per-branch semantics minerva:lint and the unit-021 drift gate use. Takes no work-unit argument; it audits the whole knowledge base.

Step 1 — Run the shape signal (deterministic, read-only)

Run migration_status through its importable Python API, anchoring both the scripts/ import path and the corpus path to the working-tree root so it works from any subdirectory (scripts/migration_status.py is read-only — it never writes):

ROOT="$(git rev-parse --show-toplevel)"; PLUGIN_SCRIPTS=$(find -L "${HOME}/.claude/plugins/minerva" "${HOME}/.claude/plugins/cache/agent-marketplace/minerva" -maxdepth 2 -type d -name "scripts" 2>/dev/null | head -1); python3 -c "import sys, json; sys.path.insert(0, '${PLUGIN_SCRIPTS:-$ROOT/scripts}'); \
from migration_status import migration_status; \
print(json.dumps(migration_status('$ROOT/.minerva/knowledge'), indent=2))"

The returned dict carries plain-primitive signals:

• non_conforming_files — *.md files in the knowledge dir that do not match ENTRY_RE (excluding the reserved index.md / overview.md). The migration-unique signal — these are the files invisible to all other wiki tooling.
• index_present / overview_present — booleans for the index catalog and the synthesis overview.
• entries_without_related — conforming entries whose ## Related block is absent or empty (no cross-ref edges). Derived fence-aware from the frozen detector's parse_entry (a fenced ## Related example does not count), and robust to malformed legacy entries (a conforming-named file with no **Type** / no sections is counted, never crashed on).
• conforming_entry_count — how many entries already match the convention.

Step 2 — Emit the migration checklist (read-only)

Translate the signal into a checklist. For each gap, name the existing skill that closes it and a one-line effect — do not run it, and do not reproduce its findings:

• non_conforming_files non-empty → these files must be renamed to NNN-type-slug.md (e.g. decision / bug / pattern / constraint) so the tooling can see them. ⚠️ Not automated — rename by hand (renames must update every [[…]] wikilink + the index catalog — deliberate mutation this read-only check never performs).
• index_present false → run minerva:init (it scaffolds / backfills index.md from the existing entries).
• overview_present false → run minerva:synthesize (it creates the theme-grouped overview.md).
• entries_without_related non-empty → these entries need ## Related cross-refs authored. ⚠️ Not automated — authoring which entries relate (and the relationship label) is LLM judgment; do it by hand. minerva:lint-fix only repairs reciprocals of links that already exist, not the initial edges.
• Always, after the corpus conforms → run minerva:lint to surface mechanical drift (index watermark, broken links, missing reciprocals) and minerva:lint-fix to repair the deterministic subset. Then run minerva:synthesize to (re)build the overview.

Present it as a numbered checklist with the counts from the signal, then stop. If non_conforming_files and entries_without_related are both empty and index + overview are present, report: migration: corpus is in conforming shape (<N> entries) — run minerva:lint + minerva:synthesize to confirm health.

Out of scope

• Any file mutation — no renames, no cross-ref authoring, no index/overview writes. This skill reports; the gated/judged remediation lives in minerva:init / minerva:lint-fix / minerva:synthesize (run by the user) or in future migration-APPLY units.
• Editing the frozen detector (scripts/knowledge_lint.py) — migration_status consumes its ENTRY_RE / parse_entry API, never re-derives it.
• Being a health check — mechanical drift and synthesis staleness are owned by minerva:lint and minerva:synthesize; this skill checks shape, not health.
• Scanning subdirectories — like the detector / fixer / synthesis tooling, migration_status globs *.md non-recursively (top-level of .minerva/knowledge/ only); entries nested in a subdirectory are not inventoried (consistent with the rest of the toolchain, which is equally non-recursive).