The Agent Skills Directory

[COMMAND_EXECUTION]: Executes the paseo reload command to apply changes to the environment without a full restart.
[PRIVILEGE_ESCALATION]: Automatically configures agent profiles with high-privilege access modes (full-access for Codex models and bypassPermissions for Claude models) by default.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from existing configuration files, creating a potential surface for indirect injection.
Ingestion points: Reads existing agent profiles and settings from $PASEO_HOME/config.json via the list_profiles tool.
Boundary markers: Absent; there are no specific instructions or delimiters used to separate user data from system instructions during the loading process.
Capability inventory: The skill has the ability to write to sensitive configuration files and execute shell commands (paseo reload).
Sanitization: Absent; the skill does not explicitly validate or sanitize the content of the profiles before migrating or rewriting them.
[CREDENTIALS_UNSAFE]: Interacts directly with configuration files that are explicitly stated to contain secrets. While it instructs the agent to preserve these secrets, the capability to modify these files presents a security risk if the instructions are bypassed.

setup-pstack