SET

What it does

• Resolves a preset into a small, explicit SET run plan for a repo.
• Runs thin agentsgen flows such as init, pack, check, repomap, snippets, analyze, meta, and optional proof-loop artifacts.
• Runs repo-local ID integration hooks for compatible repositories: pre_task before orchestration and weekly_review after it.
• Exports reviewable planning payloads so registry config and repo workflows stay auditable.
• Exports orchestrator-bundle.json for external agent runners that need stable repo context before spawning agents.
• Exports proposal-lifecycle.json for proposal-first flows: run, retained output, inspect, then select/apply/discard.
• Exports review-only hints for memory capability contracts, research diversity, context budgets, context degradation review, and loop readiness, plus an optional disabled L1/L2 loop-run contract.
• Exports a disabled bounded-orchestration profile with PLAN_DRAFT/PLAN_REVISE/PLAN_APPROVED, stable finding IDs, a five-review cap, disjoint executor ownership, explicit route evidence, and root-owned verification.
• Recommends review lenses such as assumption-excavation, pipeline-readiness-gate, confidence-fragility-review, hypothesis-diversification, context-degradation-review, agent-tool-contract-review, and loop-readiness-review.
• Exports an opt-in loop-hardening contract for measured harness stripping, immutable runtime-path sprint packets, and broken-window revalidation without automatic revert.
• Exports a disabled design-taste-review contract: marketing/editorial surfaces get an explicit design read and browser evidence, while product UI routes to Lazyweb and UX review first.
• Exports a disabled agent-operations contract for concrete agent capability cards, async operation and revalidated decision receipts, source-linked KEEP/ADAPT/ADD/REJECT deltas, trust-graded scoped memory, public/private state boundaries, and evidence-backed provider/tool routing.
• Exports a disabled skill-quality-pipeline contract for SkillOpt-style rollout evidence, bounded edits, held-out validation, rejected-edit memory, and best_skill.md export without self-editing or automatic adoption.
• Treats terminal multiplexers such as herdr as optional runner UI references: useful for pane/session visibility, blocked states, and detach/reattach, but not as SET dependencies or approval systems.
• Exports rabbithole.seed.md as an optional local review seed for human-in-the-loop plan inspection.
• Exports an optional, disabled-by-default governance contract for shadow policy decisions, append-only action receipts, and tool/cost/latency evaluation.
• Supports planning-only drift checks (`--repo-root`) and CI-safe contract outputs in JSON mode.

Quick look

One CI entrypoint and planning contract for keeping repo AI surfaces current, reviewable, and ready for governed downstream runners without becoming their runtime.

Technical definition: thin GitHub Action control-plane for repo-docs, repomap, proof loops, registry-driven flows, handoff bundles, optional shadow-first governance, bounded orchestration, loop hardening, design review, agent-operations, and skill-quality contracts.