The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill invokes npx @claude-flow/cli across multiple commands and scripts. Using npx with an unpinned external package from the public registry downloads and executes code dynamically at runtime, introducing a package supply chain risk.
[COMMAND_EXECUTION]: The skill relies on direct execution of shell commands and custom shell scripts (scripts/memory-backup.sh and scripts/memory-consolidate.sh) to orchestrate its core memory backup, statistics, and optimization workflows.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: Untrusted external inputs or conversational history enter the agent context dynamically via the memory store and memory search commands inside SKILL.md.
Boundary markers: There are no explicit boundary markers, escaping rules, or system instructions defined to separate untrusted text inputs from structural prompt logic.
Capability inventory: The skill possesses shell execution capabilities across all files through the use of the @claude-flow/cli tool and shell scripts.
Sanitization: Input content is stored directly into the namespace fields without active sanitization or verification, allowing a potentially malicious instruction stored in memory to execute when later retrieved during semantic search operations.

memory-management