The Agent Skills Directory

[COMMAND_EXECUTION]: The skill explicitly mandates the use of the --yolo flag with the codex CLI tool. As described in the Required Flags section, this flag is used to "Auto-approve and bypass sandbox," which removes critical security guardrails and allows the AI to execute arbitrary code or shell commands on the host system without user confirmation.
[COMMAND_EXECUTION]: The provided execution patterns demonstrate the capability to pipe file contents (cat <file>) and directory contexts (-C <directory>) into the CLI tool while operating in a non-sandboxed, auto-approval mode. This allows the agent to perform broad file system operations without safety restrictions.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its handling of untrusted data inputs like images and external source code. (1) Ingestion points: --image flag and file content via stdin in SKILL.md. (2) Boundary markers: Absent; inputs are passed directly to the execution tool. (3) Capability inventory: codex exec (subprocess execution) and file writes via -o. (4) Sanitization: Absent; no validation or escaping is performed on external inputs. This could allow an attacker to embed instructions in an image or file that the model then executes automatically via the --yolo flag.

logic-master