railyard

Plugin railyard from marketplace novotnyllc in novotnyllc/marketplace. Commands read from the repo's manifest — 27 days ago.

Skills (13)

• audit

  Reconstruct how a run of agent work actually went (decisions, delegation, models, retries, review rounds) from the Railyard run log. Use when the user asks to audit a run, how or why it ran as it did, or for a retrospective.

• cleanup-codex

  Diagnose and recover macOS Codex app-servers under descriptor pressure — read-only inspect, two-pass recycle of a detached server (native daemon restart) or of the ChatGPT/Codex desktop app, and snapshot-bound reaping of leftover processes.

• config

  Show or change a repository's Railyard settings: turn Railyard off or on for a repository and its worktrees, set the pre-push review level (full, light, checks), stop steering toward Compound Engineering stages, or turn the merge gate off. Use when the user asks to disable, quiet, lighten, or config

• deliver

  Deliver an implementation or fix through review, merge, required release or deployment, and consumer verification unless the user sets a narrower endpoint. Use when the user explicitly asks to deliver or ship a change, or names this skill.

• doctor

  Diagnose and repair the Railyard delivery setup: plugin and skill versions across harnesses and fleet hosts, marketplace freshness, hook trust, fleet config, host reachability and enrollment, desired-state sync, and leftover test processes. Use when the user asks for doctor, a Railyard health check,

• jev

  Consult TypeSafe Jev when TYPESAFE_API_KEY is set and a model-and-effort, workflow, evidence, priority, or review-investigation choice is genuinely open between eligible options. Skip it for explicit choices, clear defaults, and small dispatches.

• model-routing

  Choose a model and reasoning effort for native subagents in Claude Code or Codex, or review an allocation on request. Use before dispatching substantive delegated work, or when changing a child's model or effort.

• oracle

  Manual Oracle consult: send a scoped prompt and files to GPT-6 Pro through the user's signed-in ChatGPT browser with the `oracle` CLI. Use when the user asks for Oracle or GPT-6 Pro, or a hard design, debugging, or review question needs that model.

• orchestrate

  Coordinate explicitly requested fleet/account allocation or delegated remote-agent work across projects, hosts, and dependencies. Use when the user requests those capabilities or names this skill. Ordinary local implementation and parallel work use native tools and subagents without this skill. A bo

• setup

  Set up Railyard on the requested host: inventory installed state, install selected workflow dependencies, and validate hooks. Use for installation, configuration, or onboarding; use doctor for an existing failure. Fleet enrollment is optional and explicit.

• thermo-nuclear-code-quality-review

  Extremely strict maintainability review for abstraction quality, giant files, and spaghetti-condition growth. Use for an explicit thermo-nuclear code quality review, deep code quality audit, especially harsh maintainability review, or a selected maintainability review of concrete structural risks.

• thermo-nuclear-review

  Deep security and correctness audit of a branch's changes. Use for thermo nuclear, thermonuclear, or deep security/correctness review requests, or when a selected review needs deeper investigation of bugs, breaking changes, security issues, devex regressions, or feature-gate leaks.

• thermos

  Run paired whole-candidate correctness/security and maintainability reviews before publication, alongside Codex review. Also use for standalone reviews. Supplies findings to the existing owner.

Similar plugins