feat(anthropic-intelligence): bootstrap KB + automation + first HTML-vs-Markdown dossier#1
Open
Klangschalen wants to merge 5 commits into
Open
feat(anthropic-intelligence): bootstrap KB + automation + first HTML-vs-Markdown dossier#1Klangschalen wants to merge 5 commits into
Klangschalen wants to merge 5 commits into
Conversation
…er profiles Adds the foundation for a continuously-updated knowledge base on Anthropic and Claude Code, structured as a subtree of klangschalen/knowledge. - README + CADENCE rationale (weekly scan, monthly synthesis) - sources.yaml: prioritized source registry (P0/P1/P2) - engineers/: profiles of Boris Cherny, Erik Schluntz, Sholto Douglas, Cat Wu, Thariq Shihipar (Anthropic) + Simon Willison (external) - topics/, monitoring/ scaffolding
…kflows GitHub Actions cron pipeline: - anthropic-weekly-scan.yml (Mon 06:00 UTC) runs fetch_sources.py over sources.yaml, writes monitoring/logs/YYYY-WW.md, opens PR - anthropic-monthly-synthesis.yml (1st 06:00 UTC) runs synthesize.py over last 4 weekly logs, updates topics/*.md via Claude API, opens PR Both workflows are manually dispatchable. Monthly synthesis requires the ANTHROPIC_API_KEY repo secret; weekly scan needs no secret. Python scripts use only feedparser + PyYAML + anthropic SDK to stay small.
… output format Synthesizes the May 2026 shift after Thariq Shihipar's (Anthropic Claude Code) "Unreasonable Effectiveness of HTML" post: - Thesis & token economics (incl. measured ~2-4× output cost) - 9 use-case categories where HTML wins, with the 20 Thariq examples - Carve-outs where Markdown still wins (READMEs, Slack, LLM-to-LLM, etc.) - Counter-arguments (Kurtis Redux, HN, security, reviewability) - Concrete recommendations for our systems: agent prompts, skill design, knowledge base, content engine, Filament dashboards - Action items mapped to klangschalen/* repos Marked status: draft — Twitter source (Thariq's original) was rate-limited during research; pending direct re-fetch on next monthly synthesis.
… W20 probe First weekly-scan probe run (executed locally with stdlib XML parsing since the sandbox couldn't build feedparser's sgmllib3k dep). Real findings: - claude-code releases feed works → verified=true (v2.1.133-2.1.139 in 5 days) - Simon Willison Atom feed works → verified=true - Anthropic news /news/rss.xml returns 404 → feed=null, marked scrape-only - Reddit /.rss returns 403 to default UA → feed=null, needs UA rotation The fetch_sources.py script in the workflow will need a UA header to make the Reddit feeds work; tracked as a follow-up.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b541a795b3
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| PRIORITY: ${{ github.event.inputs.priority || 'P2' }} | ||
| run: | | ||
| cd anthropic-intelligence | ||
| WEEK=$(date -u +%Y-%V) |
There was a problem hiding this comment.
The scanner names logs with date -u +%Y-%V, which mixes calendar year (%Y) with ISO week number (%V). On Mondays at year boundaries (for example, 2024-12-30 is ISO week 2025-01), this produces 2024-01 and can overwrite an earlier 2024-01.md log from January, causing data loss and breaking chronological ordering for monthly synthesis.
Useful? React with 👍 / 👎.
|
|
||
| by_source[source["id"]] = new_entries | ||
| # Update state — keep last 200 guids per source to bound growth. | ||
| merged = list(seen_guids) + [e["guid"] for e in new_entries] |
There was a problem hiding this comment.
The dedup state is converted to a set (seen_guids) and then back to a list before truncation, so merged[-200:] keeps an arbitrary subset of old GUIDs. Once the source has many entries, this non-deterministic retention can drop recently seen items and re-surface them as "new" in later runs, creating noisy duplicate alerts.
Useful? React with 👍 / 👎.
Eigenstaendiges HTML-Konzeptdokument als Session-Review. Geht drei Schritte zurueck und arbeitet auf: - Session-Timeline (was wurde getan) - die vier Konzept-Achsen (Repo-Struktur, Automatisierung, Quellen, Output-Form) je mit Vor-/Nachteilen, Bewertungs-Scores und Entscheidung - Sammeltabelle aller 8 getroffenen Entscheidungen mit Status-Badges (geerbt vom Branch vs. in dieser Session entschieden - sauber getrennt) - Bewertungs-Matrix der Hauptkonzepte (Einfachheit/Korrektheit/Langfristig) - drei moegliche Wege vorwaerts (Subtree dauerhaft / Subtree-Split / Merge mit engineering-principles) - 12 offene Fragen fuer den Bau in Claude Code, in 4 Bloecke gruppiert, als aufklappbare details-Elemente - Markdown-Copy-Block zur Weiterverwendung Sound-Spirit-Farben, keine externen Abhaengigkeiten, durchgehend echte Umlaute. Ablage unter concept/, da es das Architektur-Dokument der KB ist.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Bootstraps a continuously-updated knowledge base on Anthropic, Claude Code, and AI engineering best practices, structured as the
anthropic-intelligence/subtree of this repo.Three commits:
anthropic-intelligence/{README,CADENCE,sources.yaml}, engineer profiles (Boris Cherny, Erik Schluntz, Sholto Douglas, Cat Wu, Thariq Shihipar, Simon Willison), scaffolding fortopics/andmonitoring/..github/workflows/anthropic-weekly-scan.yml(Mon 06:00 UTC, no secrets) andanthropic-monthly-synthesis.yml(1st 06:00 UTC, needsANTHROPIC_API_KEY), backed byscripts/fetch_sources.py(RSS aggregation) andscripts/synthesize.py(Claude API topic re-synthesis).topics/output-formats-html-vs-markdown.mdsynthesizing the May 2026 shift after Thariq Shihipar's "Unreasonable Effectiveness of HTML" (Anthropic, Claude Code team). Token economics, 20 use-case examples, counter-arguments, and concrete recommendations mapped toklangschalen/*repos.Cadence (recommended)
anthropic-weekly-scan.yml(no secrets)anthropic-monthly-synthesis.yml(needsANTHROPIC_API_KEY)workflow_dispatchBefore merging
ANTHROPIC_API_KEYunder Settings → Secrets and variables → ActionsANTHROPIC_MODEL(defaults toclaude-opus-4-7)TODO: verifyhandles inanthropic-intelligence/engineers/*.mddraft— Twitter source was rate-limited during research)Test plan
anthropic-weekly-scanfrom the Actions tab; confirm a PR is opened with a non-emptymonitoring/logs/2026-19.mdANTHROPIC_API_KEY, manually dispatchanthropic-monthly-synthesiswithtopic=output-formats-html-vs-markdown; confirm the dossier is updated withlast_synthesized: <today>(or unchanged if no log signal)Follow-ups (not in this PR)
/new-repo-bootstrapskill inklangschalen/claude-configto streamline future repo creation + allowlist updatesdogum/html-artifactsskill pattern inklangschalen/claude-configklangschalen/agent-templatesklangschalen/engineering-principlesGenerated by Claude Code