The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The "Bootstrap Mode" grants a broad range of high-privilege permissions to the agent without manual review. This includes permissions for system utilities (chmod, systemctl), network tools (curl, ssh), and powerful shell operators (pipes |, redirects >, command chaining &&). This effectively disables the platform's security mechanism of user-mediated command approval for long-running sessions. The skill also explicitly grants permission for Bash(./*), which matches any executable in the current directory and facilitates arbitrary code execution.
[INDIRECT_PROMPT_INJECTION]: The skill scans the user's home directory for .claude/settings.local.json files and merges their contents into the global configuration. An attacker who can place a file on the user's filesystem could inject overly permissive rules that would then be automatically merged and applied globally.
Ingestion points: Files matching **/settings.local.json found via the find command in SKILL.md.
Boundary markers: None; the skill assumes all discovered settings files are legitimate and trustable.
Capability inventory: The skill uses Read, Write, and Bash tools to modify the global agent configuration at ~/.claude/settings.local.json.
Sanitization: The skill performs string deduplication and superset collapsing but does not validate or filter the security implications of the permission strings being merged.

settings-consolidator