AutoBot-AI Claude Dev Skills

AutoBot-AI-specific Claude Code skills, packaged as a plugin marketplace so every developer on the platform gets the same workflow setup with one install.

These encode the AutoBot-AI development workflow — the issue-to-merge loop, full-stack debugging, and codebase auditing — and they hardcode AutoBot's conventions (main as the PR base, release as the stable branch, autobot_shared, the deploy path, the required PR-body headings). They are the project-specific counterpart to the general, reusable-anywhere skills in Claude-Dev-Skills.

Requires an AutoBot-AI checkout. These skills reference the platform's paths and conventions; they are not meant to run against an unrelated codebase.

Copyright © 2026 mrveiss · Apache-2.0.

---

The one rule

Every skill here assumes it, so it is stated once rather than repeated in each.

Say "I don't know" — never fabricate.

A plausible-sounding guessed cause, count or verdict is the one error treated as serious. Not because guessing is impolite, but because for an agent a wrong answer is not an opinion — it executes. A person who fabricates a cause is wrong until someone checks; an agent acts on it, at machine speed, against the environment it needs to keep working. A guess produces an action, the action produces new state, and the new state is then read as evidence.

And an admission is not a closure — it is an opening. "I don't know" means "I need help with this: let's research it, guide me, let's find the answer together." It is the rewarded move because it starts the conversation that solves the problem, not because it ends one.

So the follow-through is not silence. In an interactive session, ask right then — for the guidance, the decision, or the joint dig that closes the gap. Where there is nobody to ask, leave the criterion unticked and say why, fix it in scope or file it, and never drop it. An admission that opens no dialog and produces no issue has lost the finding politely.

A stated gap is a finding. An unstated one is the defect. The same missing knowledge produces either, depending only on whether someone wrote it down.

The operative form for anything mechanical: distinguish nothing found from did not look. A check that never ran and a check that passed report the same green, and only one of them means anything.

Install

In Claude Code:

/plugin marketplace add mrveiss/AutoBot-AI-Claude-dev-skills
/plugin install autobot-dev-skills@autobot-ai-claude-dev-skills

or via the CLI:

claude plugin marketplace add mrveiss/AutoBot-AI-Claude-dev-skills
claude plugin install autobot-dev-skills@autobot-ai-claude-dev-skills

Every AutoBot developer runs the same two lines and gets the identical skill set — the point of this repo. On a new machine, they restore the whole workflow.

The skills

Issue → merge

• batch-implement — Full implement→review→merge→close→discover loop for a list of GitHub issues, with self-healing retry and per-issue verification.
• implement — End-to-end GitHub issue implementation — umbrella gate, worktree, design, code, verify, PR, CI, and the three-gate closure check
• pr — Create a pull request with pre-flight branch checks, targeting main by default
• pre-merge-validate — Validate code before merging — syntax, imports, call-site impact, tests, types, and linting
• drain — Pick and solve the backlog issues that need no decision

Review

• review — Run a PR review cycle — CI diagnosis, a three-angle finder pass, lint-only auto-fix, and the merge decision
• review-fleet — Dispatch a 10-angle parallel PR review fleet (finder agents + verifier agents) that posts only confirmed, deduplicated findings to a single PR comment

Audit

• api-wiring-audit — Audit and enforce frontend/backend API contract wiring in AutoBot-AI (or any FastAPI + SPA monorepo)
• dead-code-audit — Systematic codebase audit for unwired code — identify unregistered routers, uninvoked hooks, orphaned components, and file discovery issues

Debug

• debug-autobot — Debug any AutoBot failure across the full stack — dispatches parallel investigators per layer (Vue, FastAPI, Redis, ChromaDB, NPU, Browser, AI Stack),

Platform & session

• github-cli — Use when performing any GitHub operation — issues, PRs, comments, labels, reviews, merges, file contents, branch management, or repository queries
• session-lifecycle — Mandatory start-of-session and end-of-session protocol for every Claude Code session in this repository

Configuration

debug-autobot reads service hosts from environment variables — export them from your deployment before use, never hardcode addresses:

export REDIS_HOST=… NPU_HOST=… BROWSER_HOST=… AISTACK_HOST=…

Layout

.claude-plugin/marketplace.json            the marketplace manifest
plugins/autobot-dev-skills/
  .claude-plugin/plugin.json               the plugin manifest
  skills/<name>/SKILL.md                    one skill each

Updating

claude plugin marketplace update autobot-ai-claude-dev-skills

Relationship to the AutoBot-AI repo

The general half of this discipline lives in Claude-Dev-Skills — reusable on any project. This repo holds the half that only makes sense with AutoBot-AI. Keeping them apart means a general skill improves for everyone while a project-specific one stays where its conventions apply.