Blog Cluster Engine
Plan and execute SERP-informed, internally linked content clusters from one seed keyword.
The skill scopes artifacts to a dedicated directory, states that it does not modify other skill files, and requires user confirmation before execution after planning; it also specifies XSS-safe HTML, continuation on failures, resumability, and graceful image fallback. Points are deducted because execution can batch-call blog-write, write files, and use external search and image services, while least-privilege permissions, data-flow disclosure, indirect prompt-injection boundaries, and per-operation confirmation are not fully specified here; dependency and installation risks are mostly documented only in shared repository material. No separate deduction is made for the unverified publisher.
Command routing, plan/execute phases, input validation, failure handling, resume behavior, placeholder-link resolution, and scorecard requirements are comparatively detailed, making the happy path structurally reproducible. Points are deducted because this is a static review with no execution of key paths; SERP overlap, WebSearch fallback, parallel analysis, pre-existing-file detection, and cross-skill Task calls lack skill-local executable test evidence and depend on implementations not supplied here. The static ceiling keeps this at 10 or below.
The target scenarios, trigger phrases, plan/execute inputs, expected artifacts, confirmation boundary, and relationships to strategy and related skills are clearly described, with Markdown and multiple platform outputs supported. Points are deducted because Chinese keyword/content support, mainland-China network reachability, and environmental fit boundaries are not addressed sufficiently; the WebSearch fallback to model reasoning explicitly reduces accuracy, and core SERP analysis depends on external search.
The SKILL.md uses progressive sections and on-demand references, with command tables, argument hints, a JSON schema, error handling, quality gates, version 1.9.1, and an MIT license; README, LICENSE, SECURITY, and pyproject add repository-level maintenance and dependency context. Points are deducted because this skill does not itself identify a maintenance owner, update path, or changelog entry, and several installation, dependency, and security assumptions rely on shared repository documentation; repository-level version and test claims do not substitute for skill-level verification.
The skill defines an end-to-end objective from seed research and semantic clustering through hub-and-spoke architecture, batch interlinked posts, and a scorecard. Its artifacts and linking rules are directly usable, and batch orchestration provides value over single-post writing. Points are deducted because static material cannot establish clustering accuracy, generated-content quality, link-injection completeness, or scorecard correctness; failures in sibling skills or external services can leave partial outputs requiring manual repair. The static ceiling keeps this at 7 or below.
The skill supplies a concrete JSON schema, linking rules, quality gates, failure statuses, and audit-scorecard fields; the README also reports repository-level CI and tests. Points are deducted because the selected path has no committed skill-specific tests, real execution outputs, or independently reproducible third-party evidence. README demonstrations and test claims are shared-repository assertions and do not adequately verify blog-cluster's key paths. The static ceiling keeps this at 5 or below.
- Execution writes many artifacts to the current working directory and calls sibling skills; confirm the plan path, output directory, and permitted external services before running it, especially in directories containing sensitive files.
- When WebSearch is unavailable, the skill falls back to model reasoning; SERP overlap, volume labels, and clustering decisions require manual review. Chinese-language support and mainland-China network reachability are not demonstrated.
- External research and project context enter the model instruction environment, leaving residual indirect prompt-injection risk; do not grant untrusted web or project-root content more tools than necessary.
- Per-post failures are skipped while the cluster continues, so execution may leave incomplete clusters, unresolved placeholders, or under-target articles; inspect the scorecard and perform human editorial review.
What does this skill do, and when should you use it?
Blog Cluster is the semantic topic-cluster planning and execution skill included in claude-blog. It expands and groups keywords using SERP overlap, search intent, and entity relationships, then designs a pillar-and-spoke architecture. It generates a JSON plan and a script-free SVG cluster map, and can call blog-write sequentially to produce posts with shared context and internal links. It fits teams that need structured content ecosystems, provided Claude Code, claude-blog, and WebSearch are available.
Reads a seed keyword or a Cluster Build Plan produced by blog-strategy; uses WebSearch to expand 30 to 50 keyword phrases and groups them by top-10 SERP overlap, intent, and entity relationships; designs a pillar page, spoke posts, and bidirectional internal links; writes cluster-plan.json and a static cluster-map.html; during execution, calls blog-write for the pillar and spokes, injects the FLOW evidence-triple requirement, replaces internal-link placeholders, and writes cluster-scorecard.md. It can optionally request hero images through nanobanana-mcp.
- A content team has one broad subject and needs a pillar page plus 5 to 15 supporting posts.
- An SEO editor wants keyword grouping based on SERP overlap rather than intuition alone.
- A team has a blog-strategy Cluster Build Plan and wants SERP validation before execution.
- An agency needs to generate a linked content ecosystem and audit status, links, keyword overlap, and failures afterward.
What are this skill's strengths and limitations?
- Combines SERP overlap, search intent, and entity mapping for keyword grouping.
- Plans the pillar, spoke posts, and reciprocal internal links together.
- Orchestrates multi-post writing with shared context and continues after an individual post failure.
- Produces a JSON plan, a JavaScript-free SVG map, and an execution scorecard.
- Requires Claude Code and the claude-blog capabilities blog-write, blog-chart, and blog-image.
- If WebSearch is unavailable, it falls back to Claude reasoning and records reduced accuracy.
- A run may involve one pillar and 5 to 15 spokes, creating more time and review overhead than single-post writing.
- FLOW evidence propagation, scoring, and several recommended audits depend on companion claude-blog skills.
How do you install this skill?
Install Claude Code first, then install claude-blog from the public repository. In Claude Code, run /plugin marketplace add AgriciDaniel/claude-blog, followed by /plugin install claude-blog@agricidaniel-claude-blog. You may also clone https://github.com/AgriciDaniel/claude-blog.git, inspect the installer, and run it; the README does not provide a separate package for this individual skill.
How do you use this skill?
In Claude Code, run /blog cluster and choose plan or execute. Create a new plan with /blog cluster plan <seed-keyword>; import a strategy plan with /blog cluster plan --from strategy [path]; execute an existing plan with /blog cluster execute [path-to-plan]. After planning, the skill requires presenting the plan and waiting for explicit approval before execution.
How does this skill compare with similar options?
Compared with blog-strategy, Blog Cluster goes beyond planning by performing SERP-based semantic clustering and executing the posts. Compared with blog-write, it orchestrates multiple writing calls with shared cluster context and internal linking. Compared with blog-cannibalization, it aims to prevent keyword conflict during planning, while that skill primarily diagnoses existing content.