Code Assistants

Skills and agents for AI-assisted development workflows — plan, implement, commit, PR, and monitor.

Plugins

• autopilot — plan, implement, commit, PR, and monitor
  • Upgrade notes

Installation

/plugin marketplace add awinogradov/code-assistants
/plugin install autopilot@code-assistants

See the plugin README for scoping options and local development.

Repository docs

The docs/ guides are numbered chapters in reading order — start at chapter 1 and read on, or jump to the chapter you need.

#	Document	What it covers
1	Workspace structure	where new actions, packages, and apps go, and how they wire into Turbo
2	agents field spec	how skills detect a repository's tech stack from package.json
3	Review run-summary footer	how code-review-action surfaces per-run metrics in a footer, plus the rare rotating review tip
4	Inline suggestions and AI-agent prompts	one-click GitHub suggestion blocks and a "Prompt for AI agents" block on each inline finding
5	Plan and run skills	how the plan and run skills resolve intent, reuse evidence, and deliver a plan or review-ready PR
6	release field spec	how release-action picks the right artifacts for npm packages, GitHub Actions, and plugins
7	Release auto-merge flow	the event-driven action that merges approved, all-green release PRs and propagates downstream
8	Upstream sync	the one-action upstream-sync aggregator and the thin upstream.yml consumers run, per-kind
9	Committed Repomix pack	the .repomix/pack.xml snapshot and its merge-triggered refresh
10	The pdf-create skill	the portable @react-pdf/renderer pipeline that renders a themed PDF, decoupled from the workspace
11	Linear tracker support	how agents.trackers opts a project into Linear and other issue trackers behind one JSON contract
12	Code review repository standards	how the review enforces a consumer repo's rfc/, docs/, and principles/ with status severity
13	The shared-rules skill	the single home for instruction blocks several skills need, and how a consumer reads one at runtime
14	The explore skill	the third on-ramp: prime a durable context brief, then take surgical fixes one at a time
15	The run-primed skill	the strict run path for a primed session: validate the brief against the checkout, or stop loudly
16	The linear-plan skill	storing a plan on a Linear ticket so it outlives the session that drafted it
17	The linear-run skill	executing a stored Linear plan verbatim: reuse a valid stored plan or draft a run-local plan
18	Portable single-source skills	one authored skills layout for Claude, Codex, Kimi, and other CLIs, and the verbatim sync (RFC-0002)
19	Skill token budget	why a skill body is charged per turn, the two rules for splitting one, and the byte-budget guard
20	The PR watcher	the packaged watcher that waits for a pull request so the model wakes only for an actionable event

Standards.

• Reference formatting (RFC-0001) — the single reference-formatting + readability standard, versioned as an RFC. As of v6 skills read it from shared-rules rather than inlining a copy; the two runtimes that cannot read a file keep an inlined copy held byte-identical by the sharedBlockSync test. As of v3 every reference (including cross-document files and sections) is a real, resolvable link, guarded by the linkResolution test; see rfc/ for the RFC convention
• Stack rule sets — Bun, Bun+React+Tailwind, NodeJS+React, NodeJS+React+Tailwind

GitHub Actions

• files-sync — sync declared files from upstream repos and open one PR with the differences
• upstream-sync — aggregate the six upstream maintenance syncs behind one action; every kind on by default, opt out per kind
• agents-rules-sync — sync the stack-appropriate rules/<stack>.md into AGENTS.md (with CLAUDE.md as a symlink to it) based on package.json agents.rules
• contributing-check — validate branch name, commit messages, and PR title against CONTRIBUTING.md
• contributing-sync — sync CONTRIBUTING.md, CODE_OF_CONDUCT.md, LICENSE.md, SECURITY.md, and the contributing workflow from upstream
• licenses-sync — sync the canonical licenses workflow from upstream in its own opt-out maintenance kind
• ban-patterns — fail the job when forbidden regex patterns match files in the working tree
• licenses-audit — detect the consumer's package manager (pnpm/npm/bun) and regenerate the license report on dependency-changing PRs, auto-commit drift on same-repo PRs, and fail fork PRs that ship stale license data
• release-action — conventional-commit-driven release pipeline for npm packages, GitHub Actions, and Claude plugins; supports a custom Anthropic host (gateway/proxy) via anthropic_base_url
• release-automerge — merge an approved, all-green release PR (^release-), driven by an event-based workflow so release-publish.yml runs without a manual click
• code-review-action — AI code review for pull requests via Claude Code, with a react mode for replying to bot mentions; supports a custom Anthropic host (gateway/proxy) via anthropic_base_url
• code-review-cost-monitor — watch the review run-summary footers for cost regressions on a schedule and open (or update) a deduplicated cost-report issue
• code-review-sync — sync the canonical AI code-review workflows (code-review.yml, code-review-cost-monitor.yml) from upstream and open one PR with the differences
• auto-label — label PRs with <scope>/<workspace-member> labels for the workspace members a diff touches, and prune orphan labels on merge
• repomix-sync — sync the repomix-pack workflow and repomix.config.json from upstream so each consumer commits its own codebase snapshot
• release-sync — sync the release pipeline workflows (release, publish, release-automerge) from upstream so each consumer runs the same release flow
• validate-actions — lint changed workflow files and composite action inline shell on pull requests
• perf-report-action — build a target, measure bundle sizes and Lighthouse headlines, compare against a default-branch baseline, and post a sticky PR comment with classified deltas

Contributing

• Contributing guide
• Code of Conduct
• Security Policy

License

MIT © @awinogradov