AGENT SKILL

decision-playground

List the installed decision pattern packs and run their offline recorded fixtures through the production decision evaluator

License 许可
MIT
所属仓库
jmagly/aiwg
最近更新
2026-11-12
Stars
★ 229
Forks
36

01 前言 Front Matter

name
decision-playground
script
{ "cwd": "project-root", "runtime": "node", "argsHint": "list | show <pack> | run <pack> [--fixture <id>] [--summary] | run-all | live-plan <pack>", "entrypoint": "scripts/decision-playground.mjs" }
ensures
[ { "offline-only": "no credential, network call, or action execution" }, { "real-runtime": "every fixture runs through evaluateDecisionRuleset with the Jev adapter over a recorded-replay transport" }, { "labelled-evidence": "receipts report offline-recorded execution and sanitized-recorded-fixture evidence, never live" } ]
requires
[ { "build": "the installed package's compiled decision runtime (dist/src/decision)" } ]
triggers
[ "decision playground", "offline decision patterns", "explore decision classification" ]
namespace
aiwg
platforms
[ "all" ]
description
List the installed decision pattern packs and run their offline recorded fixtures through the production decision evaluator

02 资源文件 2

03 SKILL.md 正文

Decision Playground

Explore the governed decision pattern packs shipped in the installed package.

  • list shows every pack, its status (supported, experimental, or unavailable), primitive, fixtures, and live limits.
  • show <pack> prints the manifest, its deterministic candidate policy, and the validation result.
  • run <pack> [--fixture <id>] evaluates one fixture. The pack's governed definitions, ruleset, and offline binding run through evaluateDecisionRuleset with the production Jev adapter. A recorded-replay transport returns sanitized recorded answers, so acceptance policies, typed-output validation, native batching, and durable receipts all execute for real. Add --summary for a compact view.
  • run-all runs every offline fixture and exits non-zero if a computed route differs from the fixture's documented expectation.
  • live-plan <pack> prints the non-executing live readiness plan and the pack's call, token, cost, attempt, and deadline limits.

The command line never runs a live probe. Live execution requires the programmatic runLiveDecisionPattern() API with explicit opt-in, approved egress, a logical credential, and a transport; see docs/decision/pattern-playground.md.

Outcomes are data. Every playground action stays unexecuted.

04 同仓库更多 Skills