The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes standard command-line interface (CLI) tools to gather track-record data for the articles.
Evidence: Uses gh pr list --search to find related pull requests and issues.
Evidence: Uses git log -- skills/<name>/ to determine the history and ship dates of the target skills.
Evidence: Executes a local notification script via ./notify -f /tmp/skill-article-notify.md to alert the operator of the generated content.
[EXTERNAL_DOWNLOADS]: The skill optionally integrates with an external AI service for visual asset generation.
Evidence: Calls mcp__higgsfield__* tools (Higgsfield) to generate a 16:9 title banner based on a visual prompt derived from the skill's thesis.
[DATA_EXPOSURE]: The skill reads from potentially sensitive internal directories to tailor the content's voice and verify its history.
Evidence: Accesses memory/logs/ and memory/MEMORY.md to extract real-world usage statistics.
Evidence: Reads soul/SOUL.md and soul/STYLE.md to adopt the intended writing style for the output.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes and summarizes external/untrusted content (other skills and logs).
Ingestion points: Reads the body of skills/<name>/SKILL.md and memory/logs/ files (SKILL.md).
Boundary markers: Not explicitly implemented in the prompt templates; however, the skill mandates that every claim must trace to a specific line in the source file, which acts as a grounding constraint.
Capability inventory: Capabilities include file system writes (output/), network-based image generation (Higgsfield), and notification delivery (./notify).
Sanitization: No explicit sanitization or escaping of the ingested skill content is mentioned before it is processed for the article.

skill-article