Agent Skills › gmickel/flow-next › flow-next-make-pr

flow-next-make-pr

GitHub

自动化生成PR摘要并创建或更新Pull Request。支持草稿/就绪状态、基础分支指定、内存写入及干运行模式，自动检测分支规范，无需额外模型调用。

plugins/flow-next/skills/flow-next-make-pr/SKILL.md gmickel/flow-next

Trigger Scenarios

需要为代码变更创建或更新Pull Request 希望自动化生成PR简报并推送

Install

npx skills add gmickel/flow-next --skill flow-next-make-pr -g -y

More Options

Non-standard path

npx skills add https://github.com/gmickel/flow-next/tree/main/plugins/flow-next/skills/flow-next-make-pr -g -y

Use without installing

npx skills use gmickel/flow-next@flow-next-make-pr

指定 Agent (Claude Code)

npx skills add gmickel/flow-next --skill flow-next-make-pr -a claude-code -g -y

安装 repo 全部 skill

npx skills add gmickel/flow-next --all -g -y

预览 repo 内 skill

npx skills add gmickel/flow-next --list

SKILL.md

Frontmatter

{
    "name": "flow-next-make-pr",
    "description": "Render a briefing from the aid artifact and open or update a PR via gh. Auto-detects the spec from the branch. Supports --draft, --ready, --base, --memory, --dry-run, --update, and mode:autonomous. Not Ralph-blocked.",
    "allowed-tools": "AskUserQuestion, Read, Bash, Grep, Glob, Write, Edit, Task",
    "user-invocable": false
}

/flow-next:make-pr

The host authors one grounded aid object; flowctl validates, stores and renders the briefing. Read workflow.md, then its reached references. No extra model call or hand-assembled sections. Invocation authorizes push and PR creation; --dry-run previews without repository writes, push, PR edits or memory writes. The opt-in html-lens.md loads only behind its config gate.

Define FLOWCTL from ${DROID_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT}}/scripts/flowctl, then <plugin-root>/scripts/flowctl (two levels above this SKILL.md), then .flow/bin/flowctl, choosing the first executable. Never assume a global install. Parse $ARGUMENTS: the positional token is SPEC_ID; reject unknown flags and missing base values. Carry these values between tool calls:

Argument	Variable / effect
`--draft`, `--ready`	`DRAFT_FORCE=draft
`--base <ref>` or `--base=<ref>`	`BASE_REF`; default empty
`--memory`	`WRITE_MEMORY=1`; default 0
`--dry-run`	`DRY_RUN=1`; default 0
`--update`	`UPDATE_MODE=1`; default 0; refresh an existing open PR
`mode:autonomous` or `FLOW_AUTONOMOUS=1`	`AUTONOMOUS=1`; default 0; never sets `RALPH`

Keep this skill inline so AskUserQuestion remains available. Resolve only missing information, one question at a time with a recommended option; use a numbered prompt if the tool is unavailable. NEED_INPUT: means ask outside Bash and rerun with the answer. Ralph/autonomous gaps hard-error instead. Ralph alone owns PR_URL= stdout and harness semantics. Draft rules live in create-and-finalize; a complete chained layer can be ready under autonomy. Never merge here. Evidence, paths and requirement attribution must be grounded in the export and receipts, with unknowns explicit rather than invented.