The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local CLI tools to verify model availability and account credentials. Evidence includes calls to claude auth status --json, codex login status, and grok models during the model probing phase.\n- [PERSISTENCE]: The skill modifies primary agent configuration files to ensure model settings persist across sessions. Evidence includes modifications to ~/.claude/CLAUDE.md (via include directive) and ~/.codex/AGENTS.md (via block mirroring).\n- [INDIRECT_PROMPT_INJECTION]: The skill processes external configuration data that could influence agent behavior. Ingestion points: Reads from ~/.claude/pstack-models.md and ../patina-mode/references/provider-dispatch.md. Boundary markers: No explicit boundary markers or instructions to ignore embedded instructions are present. Capability inventory: The skill can write to sensitive configuration directories and execute provider-specific CLI tools. Sanitization: No sanitization of the content from the referenced configuration files is performed beyond descriptor normalization.

setup-pstack