plaindev

Inspired by Caveman's no-fluff discipline, tuned for easy reading.

A coding-agent skill pack that makes AI answers clear and fast to scan. Three skills ship together:

• plaindev-reply — structured answers for everyday software work.
• plaindev-check — negative-only GitHub PR review via gh.
• plaindev-task — full ticket-to-PR workflow via Jira (Atlassian MCP) and gh.

Built for non-native English readers and people who find dense text hard to parse.

Why

Caveman is the inspiration here. It cuts ~75% of tokens with strict no-fluff discipline. plaindev borrows that discipline but optimizes for a different goal: clear, fast-to-scan answers for software work.

So plaindev makes a different trade-off:

• Plain words instead of compressed slang.
• Short sentences instead of telegraphic fragments.
• Standardized shapes so you always know where the answer is.
• Compactness is a side effect, not the goal.

Who it is for

Default AI output is full of phrasal verbs ("get rid of", "kick off", "look into"), idioms, soft hedging ("you might want to consider"), and preamble. This style is invisible to most native English speakers, but it adds parsing effort for everyone under time pressure.

plaindev drops that style by default. The output stays in plain words, with the answer first and a predictable shape. These readers benefit most:

• Developers who read English as a second language (ESL).
• People who find dense text hard to parse, including folks with ADHD.
• Anyone tired of AI preamble, hedging, and filler.

You do not need to identify with any of these. If clear, structured output beats friendly preamble for you, plaindev is for you.

Skills

Skill	Purpose	Invoke
plaindev-reply	Clear, structured assistant output	/plaindev-reply, "use plaindev"
plaindev-check	Negative-only PR review with gh	/plaindev-check, "check this PR"
plaindev-task	Ticket → branch → commits → PR (Jira + gh)	/plaindev-task, "run the task workflow"

Full rules:

• skills/plaindev/reply/SKILL.md
• skills/plaindev/check/SKILL.md
• skills/plaindev/task/SKILL.md

plaindev-task requirements: the Atlassian MCP connector authorized for Jira, and the gh CLI authenticated (gh auth login). Set a default Jira project with PLAINDEV_JIRA_PROJECT (env var) or a .plaindev/config line PLAINDEV_JIRA_PROJECT=KEY.

Before / after

Default AI output

Sure! I'd be happy to help. The reason your React component is re-rendering is most likely because you're creating a new object reference on each render cycle. When you pass an inline object as a prop, React's shallow comparison sees it as a different object every render, which triggers a re-render. I'd recommend using useMemo to memoize the object.

plaindev reply

Cause: Inline object prop creates a new reference each render. Fix: Wrap the object in useMemo.

Same fix. Faster to read in any language.

Install

Supported tools: Cursor and Claude Code. git is required.

Using Claude Desktop? Its local agent / code sessions run on Claude Code, so the Claude Code install below already covers them — the skills load from ~/.claude/skills/. The plain chat tab is different: it reads skills from the server, not disk, so add them once at claude.ai/customize/skills.

By default the installer registers both skills globally. Activate reply per session with /plaindev-reply or "use plaindev". Activate check with /plaindev-check or "check this PR".

To make plaindev reply always active in the current repo without invocation, pass --always-on.

Cursor

Global — invoke skills when needed:

curl -fsSL https://raw.githubusercontent.com/goprgo/plaindev/main/install/cursor.sh | bash

Also always-on reply for the current repo (writes .cursor/rules/plaindev-reply.mdc and local skills):

curl -fsSL https://raw.githubusercontent.com/goprgo/plaindev/main/install/cursor.sh | bash -s -- --always-on

Install paths:

• Global: ~/.cursor/skills/plaindev-reply/, ~/.cursor/skills/plaindev-check/, ~/.cursor/skills/plaindev-task/
• Local (--always-on): .cursor/skills/plaindev-reply/, .cursor/skills/plaindev-check/, .cursor/skills/plaindev-task/, .cursor/rules/plaindev-reply.mdc

Claude Code

Global — applied when description matches:

curl -fsSL https://raw.githubusercontent.com/goprgo/plaindev/main/install/claude-code.sh | bash

Also always-on reply for the current repo (injects plaindev block into ./AGENTS.md and local skills):

curl -fsSL https://raw.githubusercontent.com/goprgo/plaindev/main/install/claude-code.sh | bash -s -- --always-on

Install paths:

• Global: ~/.claude/skills/plaindev-reply/, ~/.claude/skills/plaindev-check/, ~/.claude/skills/plaindev-task/
• Local (--always-on): .claude/skills/plaindev-reply/, .claude/skills/plaindev-check/, .claude/skills/plaindev-task/, AGENTS.md block

Install everything at once

Every detected tool, global registration:

curl -fsSL https://raw.githubusercontent.com/goprgo/plaindev/main/install.sh | bash

Also always-on reply for the current repo:

curl -fsSL https://raw.githubusercontent.com/goprgo/plaindev/main/install.sh | bash -s -- --always-on

Specific tools (positional args):

curl -fsSL https://raw.githubusercontent.com/goprgo/plaindev/main/install.sh | bash -s -- cursor claude-code

Or clone and run locally

git clone https://github.com/goprgo/plaindev.git
cd plaindev
./install/cursor.sh --always-on   # any of the scripts above, same flags

Uninstall (global)

Remove global plaindev skills. Does not touch repo-local files.

curl -fsSL https://raw.githubusercontent.com/goprgo/plaindev/main/uninstall.sh | bash

One tool only:

curl -fsSL https://raw.githubusercontent.com/goprgo/plaindev/main/uninstall.sh | bash -s -- cursor

Same flag on install scripts also works: install/cursor.sh --uninstall.

Local cleanup (manual):

• .cursor/skills/plaindev/
• .cursor/rules/plaindev-reply.mdc
• .claude/skills/plaindev/
• AGENTS.md block between <!-- plaindev-begin --> and <!-- plaindev-end -->

How plaindev activates (by tool)

Tool	Global install	--always-on (this repo)
Cursor	Invoke /plaindev-reply or /plaindev-check	Reply always active via rule; check on invoke
Claude Code	Auto-discovered by description	Reply always loaded via AGENTS.md

To turn a skill off in any session, use its turn-off phrase. This does not uninstall plaindev.

Triggers

reply

• Turn on: /plaindev-reply, "plaindev mode", "use plaindev"
• Turn off (session): "stop plaindev reply", "stop plaindev"
• This response only: "explain in detail", "be thorough", "long answer"

check

• Turn on: /plaindev-check, "check this PR", "pr check"
• Turn off (session): "stop plaindev check", "stop plaindev"
• This response only: "brief check", "table only"

task (on-demand only — never auto-triggered; disable-model-invocation: true)

• Turn on: /plaindev-task, "run the task workflow", "ticket to PR"
• Turn off (session): "stop plaindev task", "stop plaindev"
• Scope one run: "no ticket", "ticket only", "draft PR", "skip review"

all

• Turn off (session): "stop plaindev"

Response shapes (reply)

plaindev reply uses four shapes. Once you learn them, scanning becomes automatic.

Kind	Shape
Bug fix	Cause → Fix → Files → Test
Feature / larger change	Plan → Files → Risks
Question	Answer → Why → Example
Pushback	Concern → Evidence → Alternative

Hard rules (short version)

• One idea per sentence. Max ~15 words.
• No phrasal verbs, idioms, hedging, filler, or pleasantries.
• Active voice. Lead with the answer.
• Headings for 2+ sections. Bullets for 3+ items.
• Code blocks and error strings stay exact.

The name

plain + dev: plain language for developers. The lowercase spelling is canonical everywhere.

Status

Hobby project. v0. Used daily by the author. Feedback and PRs welcome.

License

MIT.