The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill recommends executing code from unverified community-maintained repositories using package runners.
Evidence: npx -y @sooperset/mcp-atlassian in SKILL.md utilizes the -y flag to bypass confirmation prompts for downloading and executing third-party code.
Evidence: uvx mcp-atlassian in references/mcp-server-setup.md also performs remote fetch and execution of unverified community packages.
[INDIRECT_PROMPT_INJECTION]: The skill implements automated triage and synchronization workflows that are vulnerable to malicious instructions embedded in processed Jira tickets or Confluence pages.
Ingestion Points: jira_get_issue (fields: summary, description, changelog) and confluence_get_page (field: body.storage) as seen in references/common-workflows.md.
Boundary Markers: None detected. The scripts lack delimiters or instructions to ignore embedded commands in the ingested text.
Capability Inventory: The skill possesses extensive write and delete capabilities, including jira_update_issue, jira_add_comment, jira_delete_issue, confluence_create_page, and confluence_update_page.
Sanitization: Partial sanitization exists (escapeHtml in the release notes generator), but the core triage logic (triageNewIssue) uses direct regex matching on raw issue content, which can be manipulated by an attacker to influence routing or priorities.
[EXTERNAL_DOWNLOADS]: The skill references and encourages the use of multiple third-party tools from non-verified namespaces.
Evidence: Recommendations for atlassian-mcp (xuanxt) and mcp-atlassian (sooperset) in references/mcp-server-setup.md.
[COMMAND_EXECUTION]: The skill provides configuration templates for IDE integration that execute shell commands to start MCP servers.
Evidence: The atlassian server configuration in SKILL.md uses the npx command with environment variables that include API tokens, creating a risk if the environment is not properly secured.

atlassian-mcp