By name: wiki-daily-report description: Generate a Japanese-language daily wiki change report summarizing today's modifications. Handles the case where wiki_report.py doesn't exist by using find-based fallback. version: 1.1.0 author: Hermes Agent tags: [wiki, report, cron, japanese] category: wiki
Wiki Daily Report
Description
Generate a Japanese Discord-format daily wiki change report. Leads with high-newsjacking-signal stories (NJ ≥ 4), then lists all changes sorted by NJ score, ending with stats.
Prerequisites
- Wiki at
~/wiki/(resolves to~/ai-topics/wiki/) - Log files:
wiki/log.mdorwiki/log-2026.md
Primary Approach: Read log.md (Most Efficient)
The upstream cron jobs (newsletter-ingest, blog-ingest, active-crawl) write detailed structured entries to wiki/log.md. This is the best source of change data — far more efficient than file-system discovery.
Step 1: Extract Today's log.md Entries
# Find all section lines for today's date
grep -n '^## \[' ~/ai-topics/wiki/log.md | grep "$TODAY"
# Read the full log entry sections for today (adjust filename if rotated)
# log.md is preferred; check log-2026.md if the current year's log is rotated
Read the full log.md file — it's typically <200 lines and gives you structured entries per ingest job:
## [DATE] newsletter ingest— newsletter batch results## [DATE] blog-triage ingest— blog processing results## [DATE] Active Crawl— knowledge crawl results## [DATE] Symphony blog article ingestion— direct blog ingestion
Step 2: Identify Top NJ Stories from log.md
From the log entries, identify the most impactful stories. For each:
- Read the corresponding concept/entity page (not raw articles) — these already have structured content
- Apply NJ scoring based on the page's summary + known industry context
- Top scorer becomes the 🔥 lead; NJ 2-3 go to 📊 summary
No need to read raw articles unless the concept page is too sparse.
Step 3: Aggregate Stats from log.md Headers
Grep log.md entries for counts:
# Count new concepts — look for "Created:" or "New Concept Pages" or "Created: concepts/"
grep -c "Created: concepts/" ~/ai-topics/wiki/log.md
# Count new entities
grep -c "Created: entities/" ~/ai-topics/wiki/log.md
# Count updated — "Updated:"
grep -c "Updated:" ~/ai-topics/wiki/log.md
# Get total page count from last index.md update line
grep "total" ~/ai-topics/wiki/log.md | tail -1
Fallback: Script Not Found + No log.md Entry
If the script doesn't exist AND log.md has no entry for today (rare — happens when the upstream ingest pipeline missed a tick), use file-system discovery:
# Files modified today (adjust date as needed)
find ~/ai-topics/wiki/ -type f -name "*.md" -newermt "$TODAY" 2>/dev/null
# Raw articles
find ~/ai-topics/wiki/raw/articles/ -name "*.md" -newermt "$TODAY" 2>/dev/null
# Get timestamps for sorting
find ~/ai-topics/wiki/ -type f -name "*.md" -newermt "$TODAY" -exec stat --format='%y %n' {} \; | cut -d'.' -f1 | sort
If 50+ raw articles, prioritize reading raw articles linked from new concept pages, then the 3-5 newest by timestamp, then any named after known authors.
Newsjacking (NJ) Scoring Framework
Score each article/concept on a 0-5 scale:
- 5/5: Trending wave + contrarian take + debate potential
- 4/5: Riding viral momentum + in-group resonance
- 3/5: Pattern interrupt + novelty from trusted source
- 2/5: Standard insight, well-executed
- 1/5: Incremental update
- 0/5: Noise or link dump
Report Structure (Japanese)
🔥 Top NJ Story (only if NJ ≥ 4 exists)
[NJ: X/5] タイトル — 1行サマリー
→ [[wiki/path]]
📊 Wiki Changes Summary — NJ score order (high → low)
[NJ: X/5] タイトル — 1行サマリー
→ [[wiki/path]]
📊 Stats
- 新規概念ページ: N件
- 新規エンティティページ: N件
- 新規比較ページ: N件
- 概念ページ更新: N件
- 新規Raw記事: N件
- 総更新数: N件
- NJ≥4ストーリー: N件
If no NJ ≥ 4 articles exist, lead with 📊 Today's Wiki Updates instead of 🔥 Top NJ Story.
Edge Cases
- log.md already has the day's entry: Check
log-2026.md— the "dreaming" consolidation job often writes to log.md before this report job runs. Don't duplicate content. - Raw articles without wiki pages: Include raw articles in the report even if no concept/entity page was created yet.
- Files modified but not new: If a file was modified from a previous job's run (e.g., dreaming updated frontmatter), note it as "updated" not "new".
- Git push may fail in cron: Report git status clearly but don't retry.
Writing Quality Techniques (Gwern-inspired)
The daily report pipeline can be improved using techniques from [[concepts/llm-creative-writing]] (Gwern's methodology):
Anti-Examples Trick — Strip Generic Report Tone
After generating the first draft of a report, apply a "de-ChatGPTese" pass:
- Ask the model to identify generic/platitudinous phrases in the report draft
- Reverse those patterns — strip "notably," "importantly," "it's worth noting," "in the rapidly evolving landscape"
- Force meta-cognition: ask why the generic pattern is bad and what the specific replacement should be
Generate-Rank-Select — Report Iteration
For high-importance reports (NJ ≥ 4 lead stories):
- Generate 2-3 variants of the lead section with different framing
- Select the variant with the strongest specific claim, not the most general-sounding one
- Use the rejected variants' best specific details to cross-enrich the chosen variant
Manual of Style Enforcement
Maintain an implicit style contract for all reports:
- Japanese for Discord cron outputs, English for technical wiki content
- First sentence must be the most specific claim (not scene-setting)
- No filler transitions between bullet points
- Each bullet must advance a distinct idea — merge bullets that repeat
Pitch: Anti-Examples Quality Gate in Cron Reports
When the daily-report cron detects NJ ≥ 4 stories, it should run an additional "anti-examples quality gate" pass on the lead section to strip generic framing before delivery.
Related Skills
active-knowledge-crawl— Related wiki operationswiki-entity-enrichment-from-article— Article processing workflowconcepts/llm-creative-writing— Gwern's writing quality methodology (Anti-Examples, MoS, Generate-Rank-Select)