Loop-driven development discipline — run agents in cycles of work gated by deterministic checks until a stop condition is met. Use when work needs iteration until a measurable criterion holds, when designing retry or verification loops, when a task is too large for one context window, or when choosing between one-shot orchestration and a loop. Triggers on "loop", "iterate until", "keep going until", "run until it passes", "ralph loop", "long-running", or any moment the agent is about to retry work without a defined stop condition.
Cost-efficient model tiering — the advisor pattern (a cheap executor consults a stronger model rarely) and the orchestrator pattern (a strong model plans while cheaper workers execute). Use when choosing which model runs a session vs its subagents, when cost matters on long agentic work, or when the user mentions "Fable", "advisor", "model tiering", "token cost", "cheaper model", or asks which model should drive a workflow.
Draht command prompt wrapper for orchestrate-loop. Use when the user selects the Draht orchestrate-loop workflow or wants to run a verification-gated loop of fresh subagent iterations until a deterministic check passes.
Draht's Get Shit Done workflow — how to use /new-project, /discuss-phase, /plan-phase, /execute-phase, /verify-work, /next-milestone, /pause-work, /resume-work, /progress, /fix, /quick and the .planning/ directory structure to drive a project from idea to shipping. Use when the user asks how to plan work, structure a project, set up milestones, track progress, or wants to start using draht's workflow.
Draht's Get Shit Done workflow — how to use /new-project, /discuss-phase, /plan-phase, /execute-phase, /verify-work, /next-milestone, /pause-work, /resume-work, /progress, /fix, /quick and the .planning/ directory structure to drive a project from idea to shipping. Use when the user asks how to plan work, structure a project, set up milestones, track progress, or wants to start using draht's workflow.
Decompose work into atomic reasoning units before acting — state the logical component, validate independence, verify correctness. The discipline that opens every draht slash command. Use whenever planning, designing, refactoring, breaking down a feature, structuring a commit, or scoping work. Triggers on "plan", "decompose", "break down", "scope", "split", "atomic", "how should I structure", or any moment Codex is about to spawn many parallel changes without first carving the work into independently-verifiable units.
Four-phase systematic debugging — root cause investigation → pattern analysis → single-hypothesis testing → reproducing-test-first implementation. Use whenever investigating a bug, test failure, error, stack trace, regression, build break, or unexpected behaviour. Auto-triggers on phrases like "broken", "doesn't work", "failing", "error", "bug", "regression", "why is X", "what's wrong with". The same protocol that `/fix` enforces, available transversally.
Evidence before claims — before stating that work is done, fixed, passing, working, ready, complete, or successful, run the command that proves it and read the output. Triggers on language like "done", "fixed", "should work", "looks good", "passing", "ready to ship", "complete", or whenever Codex is about to assert positive state. Applies everywhere — feature work, bug fixes, refactors, infra, docs that depend on code state.