delete-sessions
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill is designed to perform file and directory deletions on the local system.
- The agent is instructed to remove
.jsonltranscript files and their associated sidecar directories within the~/.claude/projects/path. - Safety mechanisms include a string-matching check to identify and protect the currently running session and a mandatory manual confirmation step where the user must type the specific word "delete".
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from past session transcripts to display a summary table to the user.
- Ingestion points: In Step 3, the agent reads the content of historical
.jsonlfiles to extract the first user message as a title. - Boundary markers: There are no explicit instructions to use delimiters or ignore potential commands embedded within the session transcripts during the title extraction process.
- Capability inventory: The skill uses file reading, directory listing, and file/directory deletion capabilities.
- Sanitization: The instructions recommend trimming the extracted title to a single line, but do not provide specific sanitization or escaping protocols for the content displayed in the summary table.
Audit Metadata