The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill uses the WebSearch tool to research trends and uses this external, untrusted data to generate blog content. This creates a surface for indirect prompt injection attacks where a malicious website could influence the agent's behavior.
Ingestion points: Web search results retrieved in Step 3 are processed to select topics and generate the text of the article in Step 4.
Boundary markers: Absent. The skill does not use specific delimiters or instructions to isolate the retrieved web content, making it easier for the agent to accidentally follow instructions embedded in that content.
Capability inventory: The skill has the ability to write files to the local filesystem (Step 5, 6, and 7) and execute shell commands to commit and push those changes to a remote repository (Step 8).
Sanitization: Absent. The skill does not filter or sanitize the retrieved web content before using it to generate the final HTML output.
[COMMAND_EXECUTION]: The skill executes multiple shell commands to perform its publishing tasks. In Step 8, it runs cd, git add, git commit, and git push to synchronize local changes with the remote main branch. The skill is configured to run fully autonomously without asking for confirmation, which increases the risk if the agent's output is compromised by malicious input.

blog-inboxmate