agentic-review-orchestrationlisted

# Agentic Review Orchestration **Independent, contextless multi-agent review.** Instead of one agent reviewing its own work, spawn several fresh reviewers who never saw the implementation, brief each narrowly, and treat *agreement between them* as your strongest priority signal. Worked examples and provenance: `examples.md`. Reviewer brief template: `assets/reviewer-brief-template.md`. ## Vocabulary - **Subagent** — a separate agent instance spawned for a bounded job, with its own context window. - **Contextless reviewer** — a subagent given *only* a file list and a written brief. It never sees the implementation conversation, the other reviewers, or their output. Independence is the entire value. - **Convergence** — two or more independent reviewers flagging the same issue. The strongest signal you get. - **Triage** — you deciding which findings are real and how severe. Reviewer output is an *input to judgment*, never a verdict. - **Lens** — a posture defined by *what surface it inspects* (spec, UX, CI). - **Persona** — a posture defined by *who the reviewer pretends to be*. - **Review target** — the exact base commit, head commit, owning Issue, accepted contracts, and allowed files a reviewer may inspect. - **Provisional ledger** — local mutable memory holding unapproved candidate findings. Not an implementation plan, public documentation, or reviewer input. - **Public-safe review summary** — the consolidated, verified account placed in the PR after re