The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted content from Jira issues and Confluence pages, creating a vulnerability surface for indirect prompt injection.
Ingestion points: Jira API responses processed via curl (SKILL.md) and Confluence HTML content fetched via requests (confluence-to-md.py).
Boundary markers: Absent; the skill does not use delimiters to separate ingested external data from agent instructions.
Capability inventory: The skill can execute system commands (nix, curl, uvx) and perform network operations/file writes (confluence-to-md.py).
Sanitization: Absent; while HTML is converted to Markdown, the text content is not sanitized for malicious instructions.
[COMMAND_EXECUTION]: Executes external binaries and scripts using nix run nixpkgs#acli and uvx. It also utilizes curl for REST API interactions and python3 -c for inline data processing.
[CREDENTIALS_UNSAFE]: Documentation recommends creating "classic API tokens (NOT scoped)", which grants broader permissions than necessary and increases the impact if the token is compromised.
[DATA_EXFILTRATION]: Transmits user-provided ATLASSIAN_API_TOKEN and ATLASSIAN_EMAIL to the host specified in ATLASSIAN_SITE for authentication purposes.

atlassian