The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to download and install a plugin from an untrusted community repository (baskduf/FableCodex) using the codex plugin marketplace add command.
[REMOTE_CODE_EXECUTION]: By instructing the user or agent to download a plugin from an unknown source and then add its bin directory to the PATH, the skill enables the execution of remote code downloaded from a third-party repository.
[COMMAND_EXECUTION]: The instructions include bash commands to modify the environment and execute helper binaries (codex-fable5 status, codex-fable5 goals create, etc.) that are provided by the external plugin.
[DYNAMIC_EXECUTION]: The skill facilitates the installation of third-party binaries and their integration into the execution environment via PATH modification, allowing for the runtime execution of externally provided code.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and 'adapt' third-party prompts, 'leaked' system prompts, and external documentation, which are untrusted data sources.
Ingestion points: Processes third-party prompt guidance, 'Claude/Fable-flavored' prompt guidance, and model cards found in the workspace or provided by the user (File: SKILL.md).
Boundary markers: The skill mentions treating third-party files as 'untrusted source material', but does not provide robust delimiters for the agent when processing this data.
Capability inventory: The agent has the capability to perform network-linked plugin installations and execute shell commands (File: SKILL.md).
Sanitization: The instructions advise ignoring or rewriting content that conflicts with active rules, though this relies on the model's adherence to the instruction rather than technical enforcement.

codex-fable5