The Agent Skills Directory

[DATA_EXFILTRATION]: The skill manages model configuration in ~/.meta-o/models.json and reads session history from standard provider locations (~/.claude, ~/.codex, ~/.local/share/opencode). It includes explicit instructions in references/methodology.md to redact tokens, passwords, and PII from business framing documents before they are committed to the repository.- [COMMAND_EXECUTION]: Utility scripts mo-models.mjs and mo-posture.sh execute system commands including git, npm, and provider-specific CLIs such as claude, codex, and opencode. These operations are limited to environment diagnosis, model cataloging, and repository management. The Bash diagnostic tool uses privileged mode (-p) to isolate its execution from potentially malicious caller environment variables.- [REMOTE_CODE_EXECUTION]: The script mo-models.mjs uses dynamic loading to interface with the @anthropic-ai/claude-agent-sdk. The module resolution is scoped to the local project or global npm root, and the package name is hardcoded. This facilitates model discovery without executing arbitrary external code.- [PROMPT_INJECTION]: The skill documents an indirect prompt injection surface inherent to development workflows that ingest user requests and external web content. It mitigates this risk by advising the user to treat external content as untrusted task data and providing instructions for manual inspection and redaction of ingested text.

mo-omnigent