The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data including task specifications and code diffs which could potentially contain malicious instructions aimed at influencing agent behavior.
Ingestion points: The skill ingests task_specification and current_diff as inputs defined in SKILL.md.
Boundary markers: The instructions do not specify explicit delimiters or boundary markers to isolate untrusted data from the instruction context.
Capability inventory: The skill is configured for workspace mutation (mutatesWorkspace: true) and instructs the agent to use git diff and other file inspection tools.
Sanitization: No explicit sanitization, filtering, or validation logic for the input data is provided within the skill instructions.
[SAFE]: The primary purpose and instructions of the skill are defensive in nature. By enforcing strict scope boundaries and rejecting unrequested code modifications, the skill reduces the risk of unintended side effects or the introduction of hidden malicious code during automated editing tasks.

fable-scope-discipline