The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell scripts (fable-probe.sh, fable-exec.sh) and system commands (claude, python3, grep) to manage subagent invocation and verify model identity at runtime. This includes the use of env -u to manipulate environment variables for billing control.
[COMMAND_EXECUTION]: Internal helper scripts utilize python3 -c to execute dynamically generated Python snippets for processing JSON data. This is used specifically for managing the availability cache in ~/.claude/cache/fable-availability.json.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it interpolates raw user-provided task descriptions directly into LLM prompts without sanitization or protective boundary markers.
Ingestion points: The main task description argument is the primary ingestion point for untrusted data in SKILL.md and fable-exec.sh.
Boundary markers: No delimiters or instructions are used to separate user input from system-level instructions in the generated prompts.
Capability inventory: The skill possesses capabilities to execute shell scripts, invoke the Agent tool, and spawn the claude CLI with tool access.
Sanitization: No validation or escaping is applied to the input task string before it is passed to the subagent.

fable