claude-code-transcripts
Pass
Audited by Gen Agent Trust Hub on Jun 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes the
claude-code-transcriptsCLI tool andghCLI to automate the conversion and publishing of session logs. - [EXTERNAL_DOWNLOADS]: Instructions include installing the
claude-code-transcriptspackage from PyPI using standard package managers likeuvorpip. - [DATA_EXFILTRATION]: Facilitates the transfer of local session data from
~/.claude/projectsto GitHub Gists or local directories. It also allows fetching session data from external URLs. - [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface by processing external or local session data which may contain embedded instructions.
- Ingestion points: Reads session data from
~/.claude/projects/*/session.jsonland user-provided URLs. - Boundary markers: Not explicitly defined in the provided instructions to isolate the agent from instructions within the transcripts.
- Capability inventory: Includes shell command execution, file system access to session logs, and network operations to GitHub and the Claude API.
- Sanitization: While the tool provides HTML formatting and syntax highlighting, it does not explicitly describe measures to sanitize malicious prompts embedded in the ingested data.
Audit Metadata