The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use shell commands like cp to back up configuration files at ~/.codex/config.toml and ~/.codex/rules/default.rules before applying changes.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface because it generates security rules based on data gathered from the current session.
Ingestion points: Processes command history and session actions to identify candidates for persistent approval.
Boundary markers: Implements a 'proposal' phase where the agent must present a table of intended changes for explicit human confirmation.
Capability inventory: Direct modification of agent authorization rules and project trust levels in the local file system.
Sanitization: Instructions require the agent to use minimal allowlist patterns and prioritize fixed entry-point scripts over broad commands.
[NO_CODE]: This skill consists solely of instructions in markdown format and does not provide any scripts or compiled binaries.

watchdog-approvals