Scoped hard reasoning - architecture trade-offs, cross-module debugging, security analysis, migration risk, performance investigation. Use whenever a single hard question needs deep analysis but not a conversation, even if nobody says "delegate". One question in, one decision-grade analysis out. Read-only; never…
Local code reviewer and the designated reviewer for Codex-authored diffs. Use before any merge whenever the diff was authored by Codex (Codex never reviews its own work), Codex is unavailable, its quota window is exhausted, or the code must not leave the machine. Reviews the branch diff against its spec for…
Executes mechanical, fully-specified tasks - boilerplate, unit tests, renames, formatting, docs, small isolated fixes with clear acceptance criteria. Use whenever a task is well-defined and needs no design judgment, even if nobody says "delegate". Never use for design decisions, multi-module changes, or risk-path code.
Top-tier scoped reasoning on your highest-capability model. Use ONLY when the human explicitly authorized premium spend for a specific question in this session - never auto-delegate, never use for routine work. Typical triggers - deep-reasoner's analysis proved insufficient, or the stakes of one decision justify…
Protocol for delegating implementation, investigation or review work to Codex or to local agents (fast-worker, deep-reasoner, diff-reviewer). Use it whenever work is about to be delegated, a review is needed before a merge, a Codex command fails or hits quota, or the user asks to implement a spec, review a branch, or…
AGENTS.md instructions for jeroromano/claude-code-orchestration-template, covering agents.md, what this repository is, map, working rules and validation.
Claude Code instructions for jeroromano/claude-code-orchestration-template, covering orchestration policy, model routing, delegation, review gate (manual, branch-level) and risk paths (fill in - replace with yours).