The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends installing and executing various MCP servers via npx and uvx. While the official server from Anthropic is listed as a trusted option, the skill also promotes community-maintained packages such as @sooperset/mcp-atlassian and atlassian-mcp (by xuanxt) from non-trusted maintainers.
[COMMAND_EXECUTION]: Provides shell command templates using npx and uvx for configuring MCP servers. These commands are intended to be executed within the agent's environment to dynamically fetch and run server code from public registries (NPM and PyPI).
[INDIRECT_PROMPT_INJECTION]: The automation workflows described in references/common-workflows.md ingest untrusted data from Jira issues and Confluence pages.
Ingestion points: jira_get_issue (common-workflows.md), confluence_get_page (common-workflows.md).
Boundary markers: Explicit instructions in SKILL.md advise validation, but the provided code snippets lack strict delimiters or "ignore instructions" guards when processing issue descriptions.
Capability inventory: The skill facilitates high-privilege operations including jira_update_issue, jira_add_comment, and confluence_update_page based on parsed input.
Sanitization: Code snippets use escapeHtml for display, but logic for triage and routing relies on simple regex matches against potentially attacker-controlled issue summaries and descriptions.
[METADATA_POISONING]: Automated scanners (URLite and File Reputation) flagged the documentation site jeffallan.github.io and the file SKILL.md as malicious. While these appear to be documentation resources associated with the author's GitHub Pages, the alerts suggest the presence of patterns or scripts commonly associated with higher-risk content.

atlassian-mcp