atlassian
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFECREDENTIALS_UNSAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill accesses sensitive authentication files, including
~/.local/secrets/atlassian-tokens.jsonand configuration files in~/.config/acli/. It implements a security check inscripts/lib/atlassian.tsto ensure the token store has restricted permissions (chmod 600) and validates the API token's identity against the selected profile before making requests. - [COMMAND_EXECUTION]: The skill uses the Atlassian CLI tool (
acli) and executes bundled TypeScript scripts usingbunx tsx. These operations are central to the skill's purpose of automating Atlassian workflows. - [EXTERNAL_DOWNLOADS]: The instructions recommend downloading the Atlassian CLI from
atlassian.comusing themiseandaquatools. These are well-known, trusted sources for Atlassian tooling. - [INDIRECT_PROMPT_INJECTION]: The skill ingests external content from Jira issues, comments, and Confluence pages, creating a surface for indirect prompt injection from potentially untrusted external content.
- Ingestion points: Content is retrieved via
scripts/jira-get.ts,scripts/jira-search.ts,scripts/confluence-get.ts,scripts/confluence-search.ts, andscripts/jira-comment.ts. - Boundary markers: No explicit markers or "ignore instructions" warnings are used when processing external data in the scripts.
- Capability inventory: The skill can perform network operations (REST API calls to Atlassian), execute shell commands (
acli), and read/write local files (via script file-based input and agent tools). - Sanitization: External content is processed and returned to the agent without sanitization.
Audit Metadata