The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to download and install a third-party package @agentmemory/agentmemory via NPM globally or run it directly using npx.
[COMMAND_EXECUTION]: Encourages the execution of unpinned remote code via npx -y @agentmemory/agentmemory@latest, which bypasses version checks and confirmation prompts.
[PERSISTENCE]: Automatically modifies several AI agent configuration files to ensure the memory plugin or MCP server is loaded in every future session. These include:
~/.claude-code/plugins/agentmemory.json (Claude Code)
~/Library/Application Support/Cursor/User/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json (Cursor)
~/Library/Application Support/Claude/claude_desktop_config.json (Claude Desktop)
~/.codex/plugins/agentmemory.js (Codex CLI)
[CREDENTIALS_UNSAFE]: Accesses and writes to sensitive application data directories containing user configuration and potentially other authentication settings for popular AI coding environments.
[INDIRECT_PROMPT_INJECTION]: The skill implements an architecture for capturing and recalling data across sessions which represents an indirect injection attack surface.
Ingestion points: Captured context from onFileEdit, onCommand, onDecision, and manual memory_store calls in SKILL.md.
Boundary markers: None identified in the skill instructions to separate historical memory from current user intent.
Capability inventory: The skill exposes 51 MCP tools, including capabilities for storing, searching, and deleting memory, as well as graph traversal and code linking in SKILL.md.
Sanitization: No evidence of input validation, sanitization, or escaping of retrieved memories before they are injected into the agent's prompt context.

agentmemory-persistent-memory