Cluster Corpus by Theme
SkillAI & modelsPerforms axial-coding-style thematic clustering over the substacker corpus of published posts to surface candidate sections. Uses Braun & Clarke's six-phase thematic analysis — familiarization, initial coding, searching for themes, reviewing themes, defining themes, naming. Reads full bodies, not titles. Use when re-opening the section question. Trigger keywords — cluster, theme, axial coding, thematic analysis, candidate sections.
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the Cluster Corpus by Theme skill
What this skill tells your AI
The instructions your AI receives, as published by lyndonkl/claude in skills/cluster-corpus-by-theme/SKILL.md and read by ahel’s review.
Workflow
Per Curator run:
- [ ] Step 1: Read every post in corpus/published/** end-to-end (not just titles)
- [ ] Step 2: Extract 3-5 codes per post (concepts, methods, domains)
- [ ] Step 3: Group codes across posts by semantic similarity (axial)
- [ ] Step 4: Validate clusters — split or merge where needed
- [ ] Step 5: Report candidate clusters with membership, cohesion, outliers
Output
cluster_1:
candidate_handle: "kalshi-log"
posts: [list of slugs]
cohesion: high | medium | low
centroid_codes: [top 5 codes]
outlier_posts: [weakly-attached members]
rejected_clusters: [clusters with <3 posts]
Heuristics
- Cohesion
high: ≥5 posts, shared centroid, clear register. - Cohesion
medium: 3-4 posts or mixed register. - Cohesion
low: cluster exists but coherence is weak. - Reject any cluster with <3 posts (below the 3-post floor for real sections).
Guardrails
- Read full post bodies. Titles are marketing; bodies are the beat.
- Do not force every post into a cluster. Outliers are legitimate.
- If >30% of corpus doesn't cluster coherently (all cohesion low), emit "corpus too heterogeneous" → Curator abandons section proposals this run.
- Include rejected clusters in output — they feed
recommend-pruneand "watch" candidates. - Single-threaded. Don't race on the corpus.
Signals
- GitHub stars
- 158
- Forks
- 23
- Last commit
- Sep 2026
Advanced
- Catalog kind
- skill
- Gateway key
cluster-corpus-by-theme- Source
- github.com/lyndonkl/claude