lov-auto-context
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements logic to monitor the current session transcript for "unpersisted feedback" or "reusable" rules, which the agent then persists to long-term storage.
- Ingestion points: The skill processes the current session transcript and recent turns to extract rules, preferences, and feedback (documented in
SKILL.mdunder Manual Mode and Auto Mode sections). - Boundary markers: The instructions do not define delimiters or specific "ignore" markers to prevent the agent from mistaking embedded instructions in the transcript for legitimate user preferences.
- Capability inventory: The skill uses file-writing capabilities to create new markdown files in the project's memory directory and uses the
Edittool to modify project and globalCLAUDE.mdfiles (as specified inSKILL.md). - Sanitization: While modifications to the
CLAUDE.mdconfiguration files require a diff-preview and user confirmation, the writing of new project memory files is performed automatically without a review step. - [COMMAND_EXECUTION]: The skill instructs the agent to evaluate the session state and generate specific harness-level commands (such as
/fork,/compact, or/btw) for the user to copy and execute in the terminal environment.
Audit Metadata