The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions are purely focused on proof analysis and do not contain any instructions to bypass safety filters or override agent behavior.
[DATA_EXFILTRATION]: No network operations or sensitive file access patterns were detected. The skill operates solely on the text provided by the user in the context of proof states and error messages.
[REMOTE_CODE_EXECUTION]: There are no patterns involving the download or execution of remote code. All code examples provided in the files are for illustrative purposes and intended for the user to read or manually test in their own formal verification environment.
[COMMAND_EXECUTION]: No shell commands, subprocess spawning, or system-level operations are present in the skill files.
[INDIRECT_PROMPT_INJECTION]: While the skill is designed to ingest and analyze untrusted data (user-provided proof states and error messages), it lacks any functional capabilities (such as file writing or network access) that could be exploited via injection. The risk is limited to the generation of text, which is handled by the base model's standard guardrails.
[NO_CODE]: The skill does not include any executable scripts (.py, .js, .sh, etc.); it is comprised strictly of Markdown documentation.

proof-failure-explainer