The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to search for and execute validation and build scripts found within the repository (e.g., in package.json or a scripts/ folder). This could lead to the execution of arbitrary commands if the local environment contains malicious scripts.
Evidence: "Find the content folder, the image folder, and any validation script (package.json scripts, scripts/). You will run it at the end." (SKILL.md)
Evidence: "Run the site's validation and build checks for content (frontmatter, MDX compile, broken links). Fix what fails." (SKILL.md)
[INDIRECT_PROMPT_INJECTION]: The skill researches content from external, untrusted primary and secondary sources which could contain malicious instructions designed to influence the agent's drafting or publishing process.
Ingestion points: Step 2 of SKILL.md specifies reading from announcements, docs, papers, repos, changelogs, Hacker News, Reddit, and various forums.
Boundary markers: The instructions do not define delimiters or specific "ignore embedded instructions" warnings for the external data being processed.
Capability inventory: The skill has the capability to run local shell scripts (validation), publish content via the ship tool, and promote it via social or short tools.
Sanitization: No sanitization or filtering of the researched external content is described.
[DYNAMIC_EXECUTION]: The skill relies on running local build and validation tools that are dynamically identified based on the project's structure, which constitutes execution of code that may vary between environments.
Evidence: "Run the site's validation and build checks for content (frontmatter, MDX compile, broken links)." (SKILL.md)

blog