The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interacts with external content from Habr's API. It implements safety best practices by explicitly instructing the agent to treat all API response text as untrusted data and not as executable instructions. Mandatory Evidence Chain: 1. Ingestion points: draft and article content fetched from Habr API in scripts/habr.py. 2. Boundary markers: explicit instruction to ignore embedded instructions in SKILL.md. 3. Capability inventory: ability to save and publish articles via scripts/habr.py. 4. Sanitization: human-in-the-loop requirement for previews and confirmations.
[DATA_EXPOSURE]: Authentication is handled via a sensitive cookie jar provided through the $HABR_COOKIES environment variable. The script implements robust safeguards to protect these credentials: it strictly limits the transmission of cookies to the https://habr.com domain; it uses a custom HTTP handler that disables automatic redirects to ensure session cookies are never unintentionally sent to a third-party host; and the instructions explicitly forbid the agent from printing the cookies, passing them as arguments, or persisting them to disk.

habr