The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted session transcripts to automatically update the agent's memory and skills (e.g., in src/memu/hosts/bridging/instructions.py).
Ingestion points: Transcripts located in paths like ~/.claude/projects and ~/.cursor/projects (as specified in src/memu/hosts/*/sessions.py).
Boundary markers: The skill uses standard Markdown structure to separate content but lacks explicit delimiters or warnings to instruct the processing model to ignore instructions embedded within the logs.
Capability inventory: The system can execute shell commands, perform file writes, and make network connections.
Sanitization: It implements escaping for PowerShell using _ps_quote and uses shlex.quote for shell commands to mitigate injection risks during task automation.
[EXTERNAL_DOWNLOADS]: The skill facilitates downloading and executing scripts and templates from trusted and well-known services.
Fetches installation scripts for the uv tool from Astral's domain (astral.sh).
Fetches installation scripts for the claude CLI from Anthropic's domain (claude.ai).
Fetches installation scripts for cursor-agent from the Cursor domain (cursor.com).
Fetches instruction templates and documentation from the vendor's domain (memu.pro).
[DATA_EXFILTRATION]: The skill reports operational metadata to a vendor-controlled telemetry endpoint (api.memu.so).
Evidence: Reports include client and OS versions, host platforms, and operation metrics (e.g., success counts, latency).
Mitigations: The module (src/memu/events.py) explicitly excludes sensitive data like queries and memory content. It honors DO_NOT_TRACK and provides a configuration switch to disable reporting.
[COMMAND_EXECUTION]: The skill generates and executes shell and PowerShell commands to manage recurring session synchronization tasks (e.g., in src/memu/hosts/scheduling/windows.py).

install-memu