The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied tasks and interpolates them into a prompt template for a delegated agent, creating a surface for indirect prompt injection.
Ingestion points: The task variable in scripts/delegate-fable.sh, populated from CLI parameters or standard input.
Boundary markers: The prompt uses text labels like Task: and Rules: to delimit content, which do not provide strong isolation against adversarial instructions.
Capability inventory: Orchestrates agents via herdr, including starting agents with the --trust flag, prompting, and reading terminal buffers.
Sanitization: User input is included in the prompt without sanitization or validation.
[COMMAND_EXECUTION]: The bash script executes several herdr CLI commands to manage agent sessions, including herdr agent list, herdr agent start, herdr agent prompt, and herdr agent read.
[PRIVILEGE_ESCALATION]: The script invokes cursor-agent with the --trust flag through the herdr orchestration layer, which grants the delegated agent elevated permissions to perform actions within the workspace.
[EXTERNAL_DOWNLOADS]: The documentation provides an installation command using npx to fetch the skill package from a remote source.

delegate-fable