The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to automate the version control process for newly created skills.
Evidence: SKILL.md contains the command cd <target-repo> && git add -A && git commit -m "Add <name> skill: <one-line purpose>" && git push.
Context: This behavior is required to automate the commit and push workflow within the user's local skill repositories.
[EXTERNAL_DOWNLOADS]: The skill retrieves data from external URLs to process video content and references external guidance.
Evidence: The from-video mode utilizes the watch-video tool with a user-provided URL. It also links to the anthropics/skills repository and agentskills.io.
Context: Fetching video data is central to the workflow extraction feature, and the external links provide schema guidance.
[DATA_EXFILTRATION]: The skill transmits local project data to remote Git servers.
Evidence: The integration of git push in the scaffolding workflow.
Context: This is the expected method for publishing the generated skill files to a remote repository.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it processes untrusted content into executable instructions.
Ingestion points: Data enters through video transcripts (from-video), user-pasted text dumps (from-dump), and previous conversation history (from-chat).
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' warnings when interpolating extracted content into the new SKILL.md file.
Capability inventory: The skill has the ability to write to the file system and execute shell commands (git).
Sanitization: No explicit sanitization or filtering is mentioned for the content extracted from external sources before it is written into the new skill's instruction set.

create-skill