wrap-session
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFE
Full Analysis
- [DATA_EXPOSURE]: The skill accesses local application data at
~/.claude/projects/to retrieve raw JSONL transcripts. This is used to ensure session logs are accurate and complete, particularly for resumed sessions where context may have been compacted. - [DATA_EXPOSURE]: The skill instructs the agent to capture verbatim tool calls, arguments, and returned values in a 'Harvest' section within session logs. These logs are stored locally in
Documentation/sessions/or project-specific sandboxes and are intended for building a training corpus. - [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect injection as it processes and summarizes untrusted data from the current session history.
- Ingestion points: Reads the full session history, including tool outputs and user corrections, to extract facts and logs.
- Boundary markers: None explicitly defined in the skill instructions to delimit external content within the generated logs.
- Capability inventory: The skill performs file system reads and writes to manage memory and documentation (SKILL.md).
- Sanitization: The instructions specify that path genericization and sensitive data hygiene are handled by a separate downstream 'corpus pipeline' rather than during the writing of the initial session log.
Audit Metadata