agentsclimarketplace

Engifoundry init

Skill caoyuan-fire/engi-foundry-skill/skills/engifoundry-init

EngiFoundry / engifoundry: Engineering Foundry Skill for structured AI-assisted engineering workflows

Install
npx -y skills add caoyuan-fire/engi-foundry-skill --skill engifoundry-init

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Create or migrate the EngiFoundry project scaffold and create or modify its heavyweight-only Executor, Reviewer, and Workflow configuration through one script-driven interactive flow.

SKILL.md

5.2 KB, as published. Nobody here has run it

EngiFoundry Init

Own first initialization and every later request whose intent is to modify EngiFoundry configuration. A modification request always restarts the same complete four-question flow; do not ask which field the user wants to change. Existing configuration remains active until the final answer commits its replacement.

For an explicit migration or upgrade request, read migration.md. Migration is Agent-directed: preserve legacy content, rebuild control JSON from inspected project facts, and use the full Init interaction only when reliable preference recovery is not possible.

Start

Resolve the project root from the working directory without scanning for EngiFoundry files. Identify the current host CLI by its canonical command ID, such as codex, claude, gemini, kimi, or cursor-agent.

For first initialization, run scaffold init, then start Configurator status. The scaffold adopts an existing unconfigured-direct fallback tree only when it contains no files outside .engifoundry/packages/ and .engifoundry/artifacts/verification/; it preserves those sparse PAK and Verify records unchanged while adding the complete scaffold. Any other existing root configuration or data tree is a path collision. On a collision or scaffold failure, report the exact paths and stop. For configuration modification, do not run the scaffold; invoke Configurator status once with --init-modify or -InitModify. That flag unconditionally clears only .engifoundry/cache/configurator/ and starts a new modify flow.

Relay Protocol

Configurator JSON is the question and state authority. Do not reconstruct, localize, supplement, skip, reorder, or answer its questions.

From the first Configurator invocation until complete or cancelled, emit no model-authored conversational text. In particular, never expose planning, reasoning, intention, progress, transition, tool, or command narration such as “I should…”, “I will…”, “I’m…”, “正在……”, “接下来……”, or “已完成……”. Run every script action and custom-resolution probe silently. User-visible output is limited to the script-returned fields allowed below; this restriction applies equally to initialization and --init-modify.

  • At status: question, relay notice when present, then question.context, question.prompt, every numbered question.options, and every question.hints line exactly as returned. Wait for the user's reply.
  • Submit the complete reply unchanged with Configurator answer. All questions are single-choice except a returned kind: free-text custom-description branch.
  • At status: invalid, relay only the returned message, call status silently, and relay the same current question again. Never author an explanation or infer a corrected value.
  • At status: agent-action-required, emit nothing, apply Probe Resolution silently, and return the result through Configurator resolve.
  • At status: complete, relay only completion.lines in order and completion.message exactly as returned. Then, for first initialization, read Router silently because ./engifoundry.config.json now exists.
  • At status: cancelled, stop. Because every invocation is stateful, an interrupted conversation resumes by calling status; do not discard a valid state or overwrite configuration outside the script.

While a Configurator flow is active, its current question owns the conversation. An unrelated reply is submitted unchanged and handled as invalid input. Cancel only for an explicit user cancellation request.

Probe Resolution

For every agent-action-required custom resolution or capability probe, read probing.md in full before acting. It defines the mandatory per-interaction command --version Gate, bounded invocation rules, result fields, and the user-scoped confirmed-capability cache. The Configurator remains the state authority; do not bypass, supplement, or reorder its actions.

Commands

Run paths relative to this Skill directory and always pass the resolved project root and current CLI.

  • macOS/Linux scaffold: sh scripts/init.sh init|check --project-root <project-root>
  • Windows scaffold: powershell -ExecutionPolicy Bypass -File scripts/init.ps1 -Command init|check -ProjectRoot <project-root>
  • macOS/Linux Configurator: sh scripts/configure.sh status|answer|resolve|cancel --project-root <project-root> --current-cli <id> [--locale <locale>] [--init-modify] [--user-input <reply>] [resolution fields]
  • Windows Configurator: powershell -ExecutionPolicy Bypass -File scripts/configure.ps1 -Action status|answer|resolve|cancel -ProjectRoot <project-root> -CurrentCli <id> [-Locale <locale>] [-InitModify] [-UserInput <reply>] [resolution fields]

The script owns option discovery, numeric and free-text structural validation, branch state, four-question order, cache state, and the final atomic configuration write. Do not edit its state files or construct configuration JSON manually.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.