The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from web search results (keyword expansion and entity mapping) and incorporates this information into prompts for the downstream blog-write skill. This creates a surface for indirect prompt injection if the search results contain malicious instructions designed to influence the writing agent.
Ingestion points: WebSearch results are ingested during the Plan Phase (Step 1 and Step 4) as described in references/semantic-clustering.md to build the keyword universe and map entities.
Boundary markers: The skill uses explicit delimiters (=== CLUSTER CONTEXT ===) when injecting context into blog-write calls, as detailed in references/execution-workflow.md.
Capability inventory: The skill invokes several sub-skills (blog-write, blog-analyze, blog-schema) which perform file system operations, content generation, and network requests (via research tools). It also writes multiple JSON, HTML, and Markdown files to the local directory.
Sanitization: The skill implements robust sanitization by enforcing strict slug rules (lowercase letters, numbers, and hyphens only), rejecting absolute paths/traversal attempts (..), and escaping HTML entities in its SVG cluster map to prevent XSS.
[COMMAND_EXECUTION]: The skill automates the execution of multiple other agentic skills (blog-write) in a batch process. While it requires user confirmation before starting the execution phase and handles individual failures by stopping the batch, the automation of complex content generation workflows increases the potential impact of any single failure or injection.

blog-cluster