菜单
End-to-end repo analysis: project overview, key files, maturity assessment with expert subagents per dimension, and real issues found. Use when asked how mature a repo is, to assess or audit a codebase, or for a project health check.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | maturity-analysis |
| description | End-to-end repo analysis: project overview, key files, maturity assessment with expert subagents per dimension, and real issues found. Use when asked how mature a repo is, to assess or audit a codebase, or for a project health check. |
| allowed-tools | Agent, WebFetch, WebSearch, Bash, Read, Glob, Grep |
Performs a deep end-to-end analysis of the current repository and delivers a structured report across four sections.
Launch a single Explore subagent (or a general-purpose subagent if the Explore type is unavailable) to sweep the repository. Instruct it to read the README, entry points, main modules, and core types, then report back on:
Based on the Phase 1 sweep, identify the 8–10 most important files. For each file:
Then close with a short narrative — a compact story about how these files play together.
Launch all six subagents in parallel (single message, multiple Agent tool calls). Each subagent adopts a specific expert persona, reads the relevant source files, searches the web for reference projects using the same stack or idea, and returns a structured assessment.
Persona: Junior engineer or newly-onboarded employee trying to get productive in the codebase for the first time. Assess: README quality, inline comments, architecture docs, onboarding guides. Search for a well-documented reference project on the same stack. Rate 0–10: what exists, what's missing, highest-value next step.
Persona: Senior DevOps or platform engineer who cares about developer ergonomics. Assess: linting, formatting, git hooks, dependency management, local dev setup. Search for reference projects with strong tooling on this stack. Rate 0–10: what exists, what's missing, highest-value next step.
Persona: QA lead responsible for shipping quality code with confidence. Assess: test coverage, test types (unit / integration / e2e), risk areas with no coverage. Search for reference projects with mature test suites on this stack. Rate 0–10: what exists, what's missing, highest-value next step.
Persona: Senior engineer conducting a thorough code review. Assess: naming quality, separation of concerns within single files, file structure and separation of concerns across the project, duplication, type safety, known bugs or design smells. Search for reference projects considered clean or idiomatic on this stack. Rate 0–10: what exists, what's missing, highest-value next step.
Persona: AI practitioner who builds and maintains Claude Code agent workflows.
Assess: how well the project is set up for AI-agent development — safeguards, /skills, hooks, rules, CLAUDE.md quality. Search agentskills.io for documentation on relevant skills; use the Claude Code docs as a fallback. Identify which skills or hooks would be most valuable to add. Rate 0–10: what exists, what's missing, highest-value next step.
Persona: CI/CD specialist who has designed pipelines for production systems.
Assess: the full CI/CD build-up. Search for general best practices as well as practices specific to the detected technologies and frameworks. Identify what's mature, what's missing, structural improvements, and any current issues (e.g. slow jobs, missing cache, no deploy step). Search for reference projects with exemplary pipelines on this stack. Rate 0–10: what exists, what's missing, highest-value next step. For a dedicated release/supply-chain deep-dive, recommend the sibling skill release-audit as the follow-up.
Cover all source files — but not in your own context: for small repos (up to ~50 source files) read them directly; above that, fan out parallel subagents (one per top-level directory or package) that each read their slice and return an issue list, then merge and dedupe. For each real issue (bug, design problem, type-safety gap, missing error handling, architectural inconsistency) report:
Critical / High / Medium / LowSkip purely stylistic issues unless they cause bugs. Sort by severity descending.
Deliver the final report as Markdown with these top-level sections:
## 1. Project Overview
## 2. Most Important Files
## 3. Maturity Assessment
## 4. Issues Found
Adopt portless for stable, git-worktree-aware .localhost dev URLs following its documented best practices (pinned devDependency + portless.json + dev/dev:app script split — never a hand-rolled slug or sh -c wrapper), then wire it into Conductor via .conductor/settings.toml. Detects the stack first, wraps ONLY the JS/TS frontend dev server, and researches per-workspace isolation for backends/DB/Docker that portless can't cover. Use when asked to set up portless, fix stable dev URLs across worktrees, or make a repo Conductor-friendly for parallel agents.
Create or update GitHub issues for bug reports, feature requests, and refactor tasks using the gh CLI. Use when the user wants to file a ticket, create an issue, report a bug, request a feature, plan a refactor, or update an existing GitHub issue. Also triggers for new issue, open a ticket, file a bug, feature request, create task, or mentions wanting to track work in GitHub Issues.
Sets up, audits, or migrates release-please (Conventional-Commit versioning, changelogs, GitHub releases). Use to set up release automation, automated versioning/changelogs, or release-please; to audit, review, fix, or question an existing setup; or to switch release forms (single ↔ per-module) without breaking changelog/tag history.
Sets up a repo for Conductor end-to-end: install/setup script (automatic), selectable Run targets (with more than one, no default so nothing autostarts), and archive cleanup (automatic) — writes .conductor/settings.toml. Use when asked to set up conductor, configure a workspace, add run targets, stop conductor autostarting, make a run-button menu, or add archive/cleanup on workspace removal.
Release & supply-chain readiness audit: gathers evidence from real workflows, manifests, and repo settings, compares against a gold-standard reference stack, has an adversarial subagent try to refute the draft, and returns a prioritized action plan (P1–P3) whose fixes are delegated to the sibling setup skills. Use when asked to audit release automation, publishing security, CI or supply-chain maturity, or whether a repo is ready to release. For a whole-repo health check (docs, tests, code quality) use maturity-analysis instead.
Audits and sets up .github/dependabot.yml: recommends a grouping mode (low-noise, balanced, fine-grained) from the repo's use-case, gates setup on pinned dependency versions, wires CODEOWNERS over the deprecated reviewers key, groups security updates. Use when asked to set up, review, fix, or audit Dependabot / automated dependency updates / dependency grouping.