The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script (scripts/configure.py) to automate the creation and modification of configuration files for various AI agent harnesses (Pi, Codex, Cursor, OpenCode, Claude Code). This script operates within the specified repository and includes validation logic for model identifiers.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied model identifiers which are then used to generate harness configuration files. This attack surface is handled safely by the included configuration script, which validates inputs against strict character sets and uses safe serialization formats (JSON, TOML) for all file writes.
Ingestion points: SKILL.md (collects user input for model tiers) and scripts/configure.py (reads the generated plan file).
Boundary markers: Explicit bounded blocks (e.g., ) are used to isolate generated content within existing files like AGENTS.md and CLAUDE.md.
Capability inventory: The script has the capability to write files to various configuration directories (.agents/, .pi/, .codex/, .cursor/, .opencode/, and .claude/) and mirror skill directories.
Sanitization: The validate_plan function in scripts/configure.py uses regular expressions to restrict model slugs and reasoning levels to safe character sets, and uses json.dumps to escape strings during rendering.
[SAFE]: No suspicious network operations, obfuscation, or credential exfiltration attempts were detected. The skill follows best practices for configuration management, such as using atomic file writes and verifying model ownership before mirroring directories.

ps-setup-pstack