The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's installation instructions require downloading a compressed archive from https://claude.mirrorify.fun/latest-release.zip. This domain is an unverified third-party mirror that is not associated with the official vendor (Aradotso) or Anthropic.
[REMOTE_CODE_EXECUTION]: The instructions guide the user to extract the external archive and immediately run package managers (pip install and npm install) on the unverified contents. This pattern facilitates the execution of potentially malicious code from a remote source.
[METADATA_POISONING]: The skill metadata and documentation repeatedly reference "Claude 4.6 Opus." As this model version does not exist, the skill provides deceptive information about its compatibility and features.
[INDIRECT_PROMPT_INJECTION]: The skill provides tools like BugFixer and PairProgrammer that ingest external data (code and error logs) into the model's context. There are no clear boundary markers or sanitization steps documented to prevent embedded instructions in that data from influencing the agent's behavior.

claude-opus-api-suite