// Design and implement an agent-optimized CLI interface for any project. JSON output, structured errors, exit codes, token-efficient.
| name | robot-mode |
| description | Design and implement an agent-optimized CLI interface for any project. JSON output, structured errors, exit codes, token-efficient. |
Design and implement a "robot mode" CLI for this project, optimized for use by AI coding agents.
JSON Output: Add --json flag to every command for
machine-readable output. Stable key ordering, no omitted fields.
Quick Start: Running with no args shows help in ~100 tokens. Dense, scannable, no walls of text.
Structured Errors: Error responses include code, message, and suggestions for correction. Give 1-2 relevant correct examples in error messages showing how to do what the user likely intended.
TTY Detection: Auto-switch to JSON when output is piped
(!isatty(stdout)). Human-readable when interactive.
Exit Codes: Meaningful codes:
Token Efficient: Dense, minimal output that respects context window limits. No decorative borders or padding in JSON mode.
Error Tolerance: Be maximally flexible when the intent is clear but there is a minor syntax issue. Honor commands where intent is legible; include a note teaching correct syntax for next time.
When --json is set, output only the requested format (JSON, TOML, CSV,
etc.). No prose, no explanation outside designated fields. Strict schema
adherence.
Based on Jeffrey Emanuel's Robot-Mode Maker and CLI Error Tolerance (@doodlestein)
Multi-model code review for gastown. Reviews GitHub PRs (by URL or number) or local branches (by branch name). Spawns parallel Claude, Codex, and (optionally) Gemini reviewers with specialized prompts optimized for regression prevention, then synthesizes findings into a single maintainer-grade decision report. Use --skip-gemini for dual-model mode when Gemini quota is exhausted. Invoke with /review-pr <pr-url|number|branch>.
Multi-pass code review: correctness, security, performance, and maintainability. Each finding gets severity, location, and suggestion.
Remove telltale AI writing patterns from documentation and text. Direct, concise prose without AI artifacts.
Structured brainstorming: generate 30 ideas, critically evaluate each, distill to the best 5 with actionable implementation plans.
Multi-phase planning process: understand, decompose, design, validate, then implement. No code until the plan is approved.
Evaluate and polish README and documentation. Ensure accuracy, structure, runnable examples, then de-slopify.