wiki-digest

wiki-digest

Generate a structured digest of the wiki's current state. Unlike wiki-query (which answers a specific question) and wiki-lint (which checks structural health), wiki-digest provides a high-level synthesis: what has been compiled, how it clusters, what is missing, and what connections are worth exploring.

Think of it as the weekly editorial meeting for your knowledge base.

When to use

• User asks "what's in the wiki?", "summarize what we know", "give me an overview"
• After a batch ingest to see what the new sources added
• Periodically to guide what to ingest next
• When starting a new session and wanting to orient quickly

Implementation

plugin/scripts/digest.py produces the mechanical inputs — activity counts from log.md, inventory by type/tag, tier distribution, recently updated list, top hubs, stale custom dimensions. Use the script for these. The skill's job is to add narrative layer: theme clustering, coverage gaps, cross-cutting synthesis, suggested next actions.

# Default 7-day window
python {plugin_root}/scripts/digest.py --wiki {wiki_path}

# Longer window
python {plugin_root}/scripts/digest.py --wiki {wiki_path} --since 30d

# Focus on a single tag
python {plugin_root}/scripts/digest.py --wiki {wiki_path} --focus moe

What the script does NOT do — these stay this skill's job:

• Theme clustering — group recently updated pages by topic, identify the cluster anchor, assess depth.
• Coverage gaps — "missing synthesis pages, uncompared entities, unfiled queries" requires reading bodies and judging.
• Cross-cutting synthesis — entities spanning clusters, latent contradictions.
• Suggested next actions with specific skill invocations.

So the flow: call script → digest the JSON → read top hubs and recent pages → reason about themes and gaps → produce the narrative report.

Pre-flight (orientation guard)

Always read orientation files before running:

read_file {wiki_path}/SCHEMA.md
read_file {wiki_path}/index.md
read_file {wiki_path}/log.md

Steps

① Parse arguments:

• --since=Nd or --since=YYYY-MM-DD — scope to recent activity (default: all time)
• --focus=<topic> — narrow digest to a specific theme or tag
• --format=brief — 1-page summary; full — full analysis with all sections (default)
• --tier=<tier> — default active; also all, archived, frozen. When tiers are enabled in SCHEMA.md, the inventory/cluster/gap sections restrict to the chosen tier. Tier distribution is always reported regardless of filter.

② Activity summary (from log.md): Parse log.md entries in the time window. Count by action type:

## [YYYY-MM-DD] ingest | Article Title
## [YYYY-MM-DD] query | Question asked
## [YYYY-MM-DD] lint | N issues found

Build a timeline: what was added, when, in what order.

③ Inventory scan (from index.md + script):

• Total pages by type (entities, concepts, comparisons, queries)
• Newly created pages since --since — digest.py returns recently_created (driven by created frontmatter)
• Recently updated pages since --since — digest.py returns recently_updated (driven by updated frontmatter); these may overlap with recently_created for genuinely new pages
• Most-linked pages (hub pages — top_hubs from the script)
• Orphan pages (pages with no inbound links from index scan)

④ Theme clustering: Group pages by shared tags from SCHEMA.md taxonomy. For each cluster:

• Name the cluster (the dominant tag)
• List member pages
• Identify the "anchor page" (most linked-to within the cluster)
• Note if the cluster is well-developed (3+ pages) or sparse (1–2 pages)

⑤ Coverage gap analysis: For each cluster, check:

• Is there a concept page synthesizing the cluster? If not, flag as gap.
• Are there entities mentioned in multiple pages that lack their own page?
• Are there comparison pages where two closely related entities exist but haven't been compared?
• Are there questions in the log that were never filed as query pages?

⑥ Cross-cutting synthesis (full format only): Look across clusters for emergent connections:

• Entities that appear in 3+ clusters
• Concepts that contradict each other across clusters
• Timelines or progressions implied by multiple pages
• Open questions that span multiple areas

⑥b Tier distribution (if SCHEMA.md memory_tiers.enabled: true): Compute, for every wiki page, its tier from the driving-field date (default: published_at → fallback ingested_at). A page inherits the most recent tier across its cited sources — any active source pulls the page into active. Report:

• Counts: active: N | archived: M | frozen: K
• Pages that just crossed the archived threshold in the --since window ("aging out" list — these may deserve a refresh ingest)
• Pages pinned with tier_override: (manual overrides are respected) When tiers are disabled, skip this section entirely.

⑥c Stale custom dimensions (if SCHEMA.md declares any): For each dimension where refresh_on includes digest, scan pages and find ones whose dimension value is likely stale:

• Value still equals the initial import default
• Value hasn't been updated since a newer source was ingested for the same entity
• For type: enum fields: value is not the most recent valid option (e.g. version: 1.0 when the latest ingested version is 2.3)

Offer to re-prompt for each stale page. This is non-blocking — the user can say "skip all" and digest continues. The point is a gentle reminder, not interruption.

⑦ Suggested next actions: Based on gaps and clusters, suggest specific actions:

• "Ingest {X} to fill the gap in the {tag} cluster"
• "Create a comparison page for {A} vs {B} — they share 5 pages but no comparison"
• "Run wiki-lint — {N} orphan pages detected in the inventory"
• "The {topic} cluster has grown to 8 pages — consider writing a synthesis concept page"

Output (full format)

[Operation] wiki-digest | {scope}

━━━ Activity ({since} to {today}) ━━━
Ingests: {N} | Queries: {M} | Lints: {K}
Timeline: {chronological list of key events}

━━━ Inventory ━━━
Total pages: {N} (entities: {a}, concepts: {b}, comparisons: {c}, queries: {d})
New since {since}: {N} pages
Most-linked: [[{page1}]] ({N} links), [[{page2}]] ({M} links)
Orphans: {list or "none"}

━━━ Theme Clusters ━━━
{tag-name} ({N} pages) — anchor: [[{page}]]
  {list of pages with one-line summaries}
  Status: {well-developed | sparse | no synthesis page}

...

━━━ Coverage Gaps ━━━
• {Gap description and suggested action}
• ...

━━━ Cross-cutting Synthesis ━━━
{2–4 emergent connections or themes, with page citations}

━━━ Memory Tiers ━━━            (only if enabled in SCHEMA.md)
active:   {N} pages   (driving field: {field})
archived: {M} pages
frozen:   {K} pages
Aging out since {since}: {list of pages that just crossed active→archived}

━━━ Stale Custom Dimensions ━━━  (only if any dimension has refresh_on: digest)
{dimension-name}: {N} pages with potentially stale values
  • [[page-1]] — current: {val}, newer ingest suggests: {new-val}
  • ...
→ re-prompt? [y/n/later]

━━━ Suggested Next Actions ━━━
1. {action} → {skill invocation}
2. {action} → {skill invocation}
...

[Suggested next]
→ kata:wiki-ingest <source>  or  kata:wiki-lint

Brief format (--format=brief)

Condensed to a single block:

[wiki-digest | brief | {date}]
{N} pages · {M} clusters · {K} new since {since}
Top themes: {tag1} ({N}), {tag2} ({M}), {tag3} ({K})
Gaps: {top 2 gaps}
Suggested: {top action}