reindex

name	reindex
description	Backfills or rebuilds every hub's `_CLOSETS.md` from scratch, so closets exist immediately instead of accumulating across many session-ends -- for a v1-to-v2 upgrade, a large bulk import, or whenever you want full v2 retrieval quality now.
argument-hint	[--force] [--hub <hub-name>]
disable-model-invocation	true

Memex - Reindex

Wikilink rule: When referencing any file in markdown, always use [[filename]] wikilink format.

Walk every Tier 2 hub in _MANIFEST.md and (re)generate _CLOSETS.md for each. Session-end refreshes touched hubs incrementally; reindex covers the long tail in one pass.

Picking the right migration tool:

• /memex:resummarize. Rewrites manifest + hub summaries to v2 format.
• /memex:reindex. Rebuilds the per-hub closets index used for retrieval.

Run resummarize for summary format upgrades, reindex for closets coverage. They're independent; on a fresh v1→v2 migration, run resummarize first, then reindex.

---

Step 0: Detect workspace, acquire lock, parse arguments

Run WORKSPACE_ROOT=$(pwd) && echo "$WORKSPACE_ROOT" via Bash. Read _MANIFEST.md:

• Missing or no <!-- memex-managed marker: Tell the user "Memex isn't initialized here. Run /memex:init first." and stop.

Acquire memory/.reindex.lock and check the 24-hour cooldown in memory/.reindex-runs.log. Both follow the shared bulk-write convention. See ../consolidate/references/locking.md.

Parse $ARGUMENTS:

• --hub <name>. Limit to one hub (e.g., --hub fundraising). Useful for spot rebuilds.
• --force. Rebuild every entry from scratch, ignoring mtime checks. Also bypasses the 24-hour cooldown.

---

Step 1: Build the hub list

Read _MANIFEST.md. For each row in the Hub Map, parse the wikilink in the Hub column:

• Legacy form [[<domain>-index]]. Hub folder = parent of the index file. Closets file: <hub-folder>/_CLOSETS.md.
• Closets-only form [[<domain>/_CLOSETS]] (v2.1+). Hub folder = the parent folder of the closets file. No index file expected.

Resolve absolute paths for each hub folder and its closets / archive. Archive (if any): <hub-folder>/_CLOSETS-archive.md.

Also include memory/ as a synthetic hub. Its closets file is memory/_CLOSETS.md; its files are the Tier 1 files listed in _MANIFEST.md (status, session-log, decisions, glossary, plus any user-added Tier 1 entries). The memory hub has no archive (Tier 1 is a small fixed set, not paginated).

If --hub <name> was passed, filter to that hub. --hub memory reindexes the Tier 1 closets only.

---

Step 2: For each hub, enumerate files

Enumerate the hub's member files:

• Closets-only hub (no index file). List every .md file in the hub folder, excluding _CLOSETS.md and _CLOSETS-archive.md. The folder contents are the canonical members.
• Hub with a prose index. Read the index for any explicitly-listed files (rarely present in v2.1+). Then scan the folder for .md files. Union of the two is the canonical member list.

In both cases, also flag any files on disk that aren't yet represented in _CLOSETS.md (orphans). Reindex includes orphans by default and surfaces them in the report.

For each file:

1. Skip if missing on disk (/memex:lint reports dangling hub-table entries separately).
2. Get its mtime via stat.
3. Skip if _CLOSETS.md has a current entry (closets file's mtime more recent than the underlying file's) and --force was not passed.
4. Otherwise, queue for rebuild.

---

Step 3: Generate closets entries

For each queued file, read content and write a typed-field closets entry. The format and the eight retrieval-tuned summary rules that drive each field live with session-end:

• Closets format: ../session-end/references/closets-format.md
• Summary rules: ../session-end/references/summary-rules.md

Reindex applies them verbatim. Do not deviate.

---

Step 4: Apply pagination

For each hub, after generating entries:

1. Sort all entries (rebuilt + skipped-as-current) by underlying file mtime, most recent first.
2. Top 30 stay in _CLOSETS.md.
3. Remainder go to _CLOSETS-archive.md, prepended above existing archive content (never overwrite older archive entries).
4. Both files use marker <!-- memex-closets:1.1 -->.
5. If _CLOSETS-archive.md is created for the first time, add [[<hub>-CLOSETS-archive]] to the Tier 3 table of _MANIFEST.md.
6. After overflow, append a Recently Archived section to the primary _CLOSETS.md listing the top 5 most recently demoted file stems (one line each: [[stem]] — <subjects fragment>, last touched <date>). This keeps the archive's existence and recent contents surfaced in the always-loaded layer without bloating retrieval. Section is purely informational; field-level retrieval still uses the 30 primary entries.

If a hub has ≤ 30 files, write only _CLOSETS.md with marker <!-- memex-closets:1.0 --> and no Recently Archived section.

The memory/ hub never paginates (small fixed set). Skip steps 1-6 there; just write all entries to memory/_CLOSETS.md with marker <!-- memex-closets:1.1 -->.

---

Step 5: Verify and refresh typed-edge graph

Run ${CLAUDE_SKILL_DIR}/scripts/verify-wikilinks.py against the workspace to confirm no closets entry references a missing file. List broken entries in the report; don't auto-delete.

Then refresh the typed-edge graph (no longer rebuilt at session-end):

1. Run ${CLAUDE_SKILL_DIR}/scripts/extract-graph.py against the workspace path. It writes memory/.graph.md.
2. Surface any dangling edges (typed references to nonexistent files) in the report.

Skip if the script is unresolvable. Typed edges are opt-in.

---

Step 6: Record the run and report

Append to memory/.reindex-runs.log:

YYYY-MM-DDTHH:MM:SS  hubs=<N> files=<K> refreshed=<R> skipped=<S>  status=<ok|partial>

Clear memory/.reindex.lock. Output:

Memex Reindex Report

Hubs processed: <N>
  <hub-name>: <K> files indexed (<refreshed> refreshed, <skipped> already current)
    <P> primary entries, <A> archived (if pagination triggered)
  ...

Orphans found (in folder but not in hub table):
  <hub-name>/<file.md> -- consider adding to [[<hub>-index]]
  ...

Closets issues: [PASS / N dangling references]
  ...

Total time: <Ts>

Append a one-line entry to memory/session-log.md if Memex is in full mode (skip in compatible mode):

## YYYY-MM-DD - /memex:reindex

- Reindexed <N> hubs, <K> files. <refreshed> refreshed, <skipped> already current.
- <orphans> orphans surfaced; <issues> closet issues to resolve.

---

---

Gotchas

• Reindex reads every hub-listed file. Cheap on 50 files, several minutes on 500+. Use --hub <name> to scope when iterating.
• Reindex never edits hub indexes. Only writes _CLOSETS.md and _CLOSETS-archive.md. Orphans surface for user action, not auto-add.
• --force rebuilds every entry from scratch. Use after a closets format change, or if existing entries look wrong. Otherwise the mtime check is enough.
• Manual closets edits reset mtime. A user fixing a typo in _CLOSETS.md may cause some entries to skip on the next reindex. Pass --force if that matters.
• For granular drift detection (orphans, dangling links, contradictions), use /memex:lint. Reindex is the bulk-write counterpart to lint's read-only audit.
• For summary format upgrades, use /memex:resummarize. It operates on manifest and hub summaries, not on closets.