AGENT SKILL

bootstrap

Use when someone asks by name to bootstrap a new or empty repo; it writes many files, so don't start it unasked. Runs prompts/new-project.md end to end — reference verification, stack ADR, guardrail files, typed env, and a phased plan ending in a deployable walking skeleton.

License 许可
MIT
最近更新
2026-11-15
Stars
★ 0
Forks
0

01 前言 Front Matter

name
bootstrap
description
Use when someone asks by name to bootstrap a new or empty repo; it writes many files, so don't start it unasked. Runs prompts/new-project.md end to end — reference verification, stack ADR, guardrail files, typed env, and a phased plan ending in a deployable walking skeleton.
argument-hint
<project name, what it does and for whom, platform targets, constraints, coding agents it will use, repo visibility and GitHub plan, security contact if private>
disable-model-invocation
true

03 SKILL.md 正文

<!-- Claude Code's copy of .agents/skills/bootstrap/SKILL.md, generated by the template's skills:build script; its frontmatter differs by design. Edit the .agents/ file and carry the same change into this one. -->

Run the new-project bootstrap process, exactly as written in prompts/new-project.md. Read that file in full first — it is the authoritative instruction set; this skill only wires it up. Then:

  1. Assemble the project inputs. Parse whatever was given with the request, and map what you can onto the prompt's input fields (project name, description and target users, platform targets, constraints, stack preferences, the coding agents it will use, repository visibility and GitHub plan, and a private repository's security contact). Visibility and plan never block: read them from the host, and assume private on GitHub Free for whatever can't be read, as the prompt's inputs say. Nor does the security contact: without one, follow the prompt's fallback rather than guess. The coding agents decide what AGENTS.md is wired and sized for, so include them in the question when the request doesn't name them, and record the answer in <repo>/.agents/platforms.json; one that prompts/agent-platforms.md doesn't list is researched from its vendor's documentation, as that page says. Ask, in one question, only for what neither the request, the host, nor a safe default supplies — vague inputs produce vague plans, and the prompt permits at most one blocking question thereafter.
  2. Execute the phases in order. Phase 0 (reference verification) is mandatory and comes first; if you cannot reach the network, say verification was skipped and mark every borrowed pattern provisional, as the prompt directs.
  3. Use the skeletons for every artifact the prompt requires. They're in the skeletons directory: <repo>/prompts/skeletons/ after process:apply, <template>/templates/project/ in the template repo itself. AGENTS.skeleton.md becomes AGENTS.md, and, only for the agents the selection names, CLAUDE.skeleton.md becomes CLAUDE.md for Claude Code (and the prompt gives the one-line <repo>/.gemini/settings.json that points Gemini CLI at AGENTS.md); CONTRIBUTING.md, SECURITY.md, .env.example, and the docs/ set keep their paths: reference-verification, the ADR index and the stack ADR, the ADR template (0000-template, copied unchanged: operating rule 7 points every agent at it), product, dev-plan, dependency-policy (the release runbook comes with milestone 1's deploy pipeline), and, under .github/, the CI workflow (workflows/ci.yml), the Dependabot configuration (dependabot.yml), the ruleset (rulesets/main.json, kept even where the plan can't enforce it; its required checks are the CI workflow's job names), and, once more than one person merges, CODEOWNERS. Copy each to its real path, fill every [BRACKETED] placeholder, and delete the guidance comments; the ADR template alone stays unfilled, and is left out of the check below. The env-validation skeleton is different: it's the guide for Phase 2.6's env module, so build the module from it and don't keep a copy. A finished artifact contains neither placeholders nor guidance — prove it: LC_ALL=C grep -nE '[^]A-Za-z0-9_]\[[A-Z][^]]*\]([^([]|$)|^\[[A-Z][^]]*\]([^(:[]|$)|<!-- GUIDANCE' <each artifact> must print nothing. With Node and a template checkout at hand, also run node <template checkout>/scripts/check-placeholders.mjs <each artifact>, which catches a half-deleted guidance comment the grep can't see. If the repo already has agent rules (Cursor rules, a Copilot instructions file, …), fold them into AGENTS.md rather than keep two sets, as the prompt's Phase 2.1 says, sized for the selected agents; a CLAUDE.md imports AGENTS.md.
  4. If the new project should also use the template's interface foundation (it's a fork of the template repo), follow the rebrand workflow in the template's agent reference, docs/system/agent-reference.md (init script → palette swap → tokens rebuild → contrast re-check), before building features on top.
  5. Log the run. Add a row to prompts/run-log.md when the run starts, and fill in its outcome columns (outputs, whether references were verified) when it finishes, so an abandoned run still leaves a record. If the run exposed a gap in the prompt itself, note it in the same row. In the template repo, fix the prompt there. In a repo that got the prompts through process:apply, report the gap to the template instead and leave the delivered copy unedited: a re-apply only updates files you haven't edited (prompts/README.md, "Keeping the prompts current").
  6. Report using the prompt's "Output of this prompt" contract, labeling every claim statically reviewed / executed / fully validated.

04 同仓库更多 Skills

JoelA510
JoelA510/agent-ready-template

verify

Use when a change is about to be committed, or someone asks to verify one. Runs prompts/verify.md — the repo's fast checks (lint, typecheck, the tests the change touches, generated files rebuilt) on the tree being committed, saying what passed, what failed, and what wasn't run.

★ 0
JoelA510
JoelA510/agent-ready-template

template-update

Use when this repository should take a release of the template it was built from, or a template update pull request needs finishing. Runs prompts/template-update.md — finds the template version it has, reads the CHANGELOG to the release, brings the process files across, merges each parked .template, and does what the release's entries ask.

★ 0
JoelA510
JoelA510/agent-ready-template

root-cause

Use when something is broken and the cause isn't known. Runs prompts/investigate.md — reproduce the reported case, confirm a baseline, test hypotheses one at a time, fix the root cause, find sibling occurrences.

★ 0
JoelA510
JoelA510/agent-ready-template

review-change

Use when someone asks for a review, a critical or hard look, or a second opinion on a change before it merges, whether a PR, a branch, a diff range, or a pasted diff. Runs prompts/review.md — an adversarial, evidence-cited review, including the "passes its own tests" failures.

★ 0
JoelA510
JoelA510/agent-ready-template

pre-pr

Use when a change is ready to leave your hands — before opening or updating a pull request. Runs prompts/pre-pr.md — merge the base, run the full gate, prove new checks can fail, self-review, sweep docs, and draft the description.

★ 0
JoelA510
JoelA510/agent-ready-template

plan-change

Use when someone asks to plan a feature, a fix, or a refactor before it's built, or a change is too large to start on directly. Runs prompts/plan-change.md — sizes the change and stops for a small one; otherwise writes docs/changes/<date>-<slug>.md against docs/product.md, AGENTS.md, and the ADRs, with behavior, design, risks, and tasks that each name their test.

★ 0