The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill references and operates on sensitive configuration directories in the user home directory which are used to define agent behavior and permissions.
Evidence: The skill targets paths such as ~/.claude/skills/, ~/.agents/skills/, ~/.grok/rules/, and $COPILOT_HOME/skills/ for installation and maintenance tasks.
[PERSISTENCE]: The skill modifies configuration files that are automatically loaded by AI agents, establishing a persistent presence in the agent's instruction set.
Evidence: It manages markered instruction blocks in files like CLAUDE.md and AGENTS.md, which are typically loaded at the start of an agent session to define its rules and capabilities.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a mechanism for injecting instruction snippets into files that the agent processes as instructions, creating a potential attack surface.
Ingestion points: Instructions are injected into CLAUDE.md, AGENTS.md, and various SKILL.md files.
Boundary markers: The skill uses HTML-style comment markers and to delimit its managed content.
Capability inventory: The skill leverages the agent's file-writing tools to modify files in project-local and global directories (~/.claude, ~/.agents, etc.).
Sanitization: There is no mention of sanitization or validation of the instruction payloads provided by the memory_install_self_routing tool before they are written to the filesystem.

ai-memory-routing-install