The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses and uploads Claude Code session history files from the local directory ~/.claude/projects/ to the vendor's cloud service. This is the primary and documented purpose of the skill.
[COMMAND_EXECUTION]: The skill executes vendor-provided CLI commands, specifically leverage whoami and leverage sessions import, to facilitate identity checks and the transfer of session data.
[EXTERNAL_DOWNLOADS]: The instructions suggest using npx to install additional skills from the vendor's repository (leverage-computer/skills).
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill reads and processes Claude session history files (.jsonl) which may contain untrusted content from previous LLM interactions.
Boundary markers: No specific boundary markers or 'ignore' instructions are provided to the agent to mitigate potential injection risks embedded within the history data.
Capability inventory: The skill includes capabilities to read local files and perform network uploads via the leverage CLI.
Sanitization: No explicit sanitization or validation of the session history content is performed by the skill instructions, as the data is intended for direct import into another session.

agent-handoff