mimic
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses a subagent to execute shell commands such as
grep,sed,git, andgh. These tools extract the user's writing style from session logs and repository history to create a stylistic profile. - [DATA_EXFILTRATION]: The skill reads potentially sensitive local files, including session logs from
~/.claude/projects/and~/.codex/sessions. These files contain the user's previous prompts and tool interactions, which are accessed to analyze writing patterns. - [INDIRECT_PROMPT_INJECTION]: The skill analyzes messages from external platforms like Slack and GitHub to mimic tone. It lacks robust boundary markers to differentiate between message content and potential instructions, creating a surface for indirect prompt injection.
- Ingestion points: Target messages, Slack threads, and GitHub comments (SKILL.md Step 1, Setup Step 1).
- Boundary markers: No specific delimiters or "ignore instructions" warnings are applied to the analyzed external data.
- Capability inventory: Shell command execution (grep, sed, git, gh) and local file writing (~/.mimic.md).
- Sanitization: The skill instructs the subagent to scrub secrets and tokens from the samples but does not include sanitization for embedded instructions.
Audit Metadata