atmos-asciicast
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes terminal recording files (.cast) which represent untrusted external data captured from shell sessions. \n
- Ingestion points: File SKILL.md describes reading .cast files for validation using load_text and fs.read_file functions. \n
- Boundary markers: Starlark validation scripts act as functional boundaries, and instructions advise using raw text for ANSI assertions vs stripped text for content. \n
- Capability inventory: The skill uses type: shell, type: cast, and Starlark interpreters (SKILL.md) which allow command execution and filesystem interaction. \n
- Sanitization: Instructions mandate stripping ANSI, removing secrets, local paths, and hostnames from recordings before commitment. \n- [DYNAMIC_EXECUTION]: Workflows utilize the Starlark interpreter (interpreter: starlark) and shell environments (type: shell) for task automation and content verification as described in SKILL.md. \n- [COMMAND_EXECUTION]: The workflow orchestrates the atmos CLI via SKILL.md instructions to perform setup, recording, and rendering tasks (e.g., atmos casts setup, atmos cast render).
Audit Metadata