The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is centered around the use of the codex CLI tool to execute complex tasks. It instructs the agent to use codex exec for code implementation, refactors, and codebase analysis. This involves executing system-level actions via a CLI tool based on instructions generated at runtime.
[PROMPT_INJECTION]: The skill creates an indirect prompt injection surface through its core delegation mechanism. The "Fable" agent takes user input and uses it to construct "delegation prompts" for the codex tool. If a user provides a malicious request, it could influence the commands executed by the Codex sub-process.
Ingestion points: User instructions are ingested in the SKILL.md "Think" and "Spec" phases to generate delegation plans.
Boundary markers: The skill lacks explicit instructions to wrap or delimit untrusted user content within the generated delegation prompts.
Capability inventory: The codex CLI tool has high capabilities for file system modification and codebase interaction.
Sanitization: There are no defined mechanisms for sanitizing or escaping user-influenced data before it is passed to the execution tools.

use-fable