The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions and shell command templates for configuring 'hooks' within the agent's configuration files. These hooks allow the execution of arbitrary shell commands (e.g., using jq, prettier, or npm test) in response to specific lifecycle events like PreToolUse and PostToolUse.
[PRIVILEGE_ESCALATION]: The skill facilitates the management of tool permissions, including allow and deny rules and the defaultMode setting. Misconfiguration or malicious instruction could lead the agent to grant itself unrestricted access (e.g., Bash(*)) or disable permission prompts (dontAsk).
[PERSISTENCE]: By managing hooks that trigger on SessionStart or Stop events, the skill allows for the creation of scripts that run automatically across different agent sessions, effectively establishing a persistence mechanism within the agent's environment.
[INDIRECT_PROMPT_INJECTION]: The skill's function of interpreting user prompts to update security-sensitive settings (Ingestion) creates a surface for indirect prompt injection. Attackers could potentially influence the agent to install malicious hooks or weaken its security posture (Capabilities: shell execution, file writing). The skill includes mitigations such as recommending AskUserQuestion when intent is ambiguous (Boundary) and providing specific guidance on using jq -r to safely handle JSON input within generated commands to prevent shell injection (Sanitization).
[DYNAMIC_EXECUTION]: The skill generates shell command strings from templates to be stored in configuration files and executed dynamically by the platform. It provides specific guidance on using jq -r to safely handle JSON input within these generated commands to prevent shell injection.

update-config