Short Circuit

Short Circuit

Operate like an implementation owner who cuts low-signal paths early, keeps work bounded, and uses extra agentic machinery only when it buys real quality. The instructions are Codex-first, but the core workflow is portable to other coding agents.

Reference Map

• Read references/ORCHESTRATION.md for planning, model/agent selection, N-phase execution, fresh-context judgement, codex exec, and hook usage.
• Read references/PRODUCT_INTENT.md when defining feature purpose, north-star, non-goals, docs/examples as product surface, release readiness, or product-owner behavior.
• Read references/ABSTRACTION_BOUNDARIES.md when deciding between generic primitives and case-specific patches, weighting external references, controlling scope, or simplifying public APIs.
• Read references/REPO_STRUCTURE.md when creating, reorganizing, or reviewing Python repo layout, pyproject.toml, Makefile, CI, docs, examples, scripts, generated files, or monorepo package structure.
• Read references/PYTHON_ARCHITECTURE.md when designing Python modules, APIs, data structures, adapter interfaces, decorators, named constructors, async boundaries, or error models.
• Read references/ENGINEERING_QUALITY.md when writing or reviewing code quality, typing, tests, lint/type gates, coverage, cleanup, and release hygiene.
• Read references/DX_AND_SEMANTICS.md when working on semantic contracts, schemas, metrics, CLI/YAML/LSP/traceback/docs surfaces, or developer experience.
• Read references/EVAL_AND_OPTIMIZATION.md when designing evaluation, benchmark, optimizer, trace, feedback, scoring, or judge workflows.
• Read references/AGENTIC_ENGINEERING.md when working on model backends, tool systems, run context, message/event schemas, structured outputs, retry/approval/deferred flows, or agent tests.
• Read references/SCRIPTS.md when a bundled script can make repo auditing or isolated Python execution more deterministic.

Default Behavior

When this skill triggers:

1. Identify which reference files are relevant to the task.
2. Read only the relevant references before planning or acting.
3. Use extra agents, hooks, or nested Codex runs only when they reduce real risk more than they add complexity.
4. Keep the main agent responsible for final judgement and implementation decisions.

When guidance conflicts, resolve in this order:

1. Direct user instruction.
2. Existing repo conventions, constraints, and committed configuration.
3. Approved product intent, system design plan, and implementation plan.
4. This skill's defaults and templates.

Do not force a template shape over a working repo convention without a reason.

Production boundary:

This skill improves engineering and orchestration quality, but it does not by itself prove production readiness for live services. When work involves credentials, secrets, user data, PII, external traffic, deployment, rollback, retention, or abuse/security risk, require explicit hardening evidence beyond this skill's normal coding gates.

Host equivalents:

• Codex fresh run: codex exec; other agents: a read-only subagent, separate task, or new session that writes a report file.
• Codex model controls: /model, CLI flags, or custom agents; other agents: the host's model picker or planner/implementer tiers.
• Hooks: Codex hooks when available; otherwise deterministic local automation such as pre-commit, CI, or scripts.