Maxed Apps Agent Skills

Available skills

• awesome-tests — plan, write, improve, and review behavior-focused automated tests and test strategies.
• code-review — perform adaptable generic, scoped, and plan-backed implementation reviews.
• create-changes-report — create an interactive HTML review artifact after repository changes.
• decomplex — prevent, audit, and triage unnecessary complexity without editing reviewed targets.
• explain — manually produce context-grounded Markdown explanations and standalone HTML that opens in the default browser unless suppressed.
• create-plan — create, review, and improve researched, implementation-ready plans before coding.
• create-report — create standalone HTML decision, assessment, research, and status reports.
• create-skill — create, rewrite, and review concise, actionable Agent Skills.
• create-slides — build, verify and export templated HTML slide decks from five looks (PDF + MP4 at 1080p/2K/4K).
• create-wiki — create and update reusable, source-backed technology knowledge bases.
• generate-image — generate AI images through fal.ai HTTP queue workflows (Bun CLI; default openai/gpt-image-2).
• implement-plan — execute existing Markdown implementation plans with delegation-first tracking, verification, and a final changes report.
• use-worktrees — create, sync, merge, and clean up isolated Git worktrees safely.
• use-subagents — portable delegation policy, assignment contracts, worktree isolation/cleanup for any harness.
• use-pi-subagents — Pi RPC launcher for bounded subagents when native subagent_* tools are inactive (use with use-subagents).
• use-mcp — safely discover and call targeted MCP servers through mcporter with explicit authentication gates.
• web-research — perform current, source-backed research across web content and repositories.
• vps-setup-hardening — manually set up and harden Ubuntu 24.04/26.04 or AL2023 on EC2.

Install

List all available skills:

npx skills add maxedapps/agent-skills --list

Install one skill:

npx skills add maxedapps/agent-skills@awesome-tests
npx skills add maxedapps/agent-skills@code-review
npx skills add maxedapps/agent-skills@create-changes-report
npx skills add maxedapps/agent-skills@create-report
npx skills add maxedapps/agent-skills@create-wiki
npx skills add maxedapps/agent-skills@use-mcp
npx skills add maxedapps/agent-skills@use-worktrees
npx skills add maxedapps/agent-skills@explain
npx skills add maxedapps/agent-skills@vps-setup-hardening

Or use the explicit option:

npx skills add maxedapps/agent-skills --skill awesome-tests
npx skills add maxedapps/agent-skills --skill code-review
npx skills add maxedapps/agent-skills --skill create-changes-report
npx skills add maxedapps/agent-skills --skill create-report
npx skills add maxedapps/agent-skills --skill create-wiki
npx skills add maxedapps/agent-skills --skill use-mcp
npx skills add maxedapps/agent-skills --skill use-worktrees
npx skills add maxedapps/agent-skills --skill explain
npx skills add maxedapps/agent-skills --skill vps-setup-hardening

Install all eighteen skills explicitly:

npx skills add maxedapps/agent-skills \
  --skill awesome-tests \
  --skill code-review \
  --skill create-changes-report \
  --skill create-report \
  --skill decomplex \
  --skill explain \
  --skill create-plan \
  --skill create-skill \
  --skill create-slides \
  --skill create-wiki \
  --skill generate-image \
  --skill implement-plan \
  --skill use-worktrees \
  --skill use-subagents \
  --skill use-pi-subagents \
  --skill use-mcp \
  --skill web-research \
  --skill vps-setup-hardening

Review each skill and its compatibility requirements before use. code-review, create-plan, and implement-plan delegate by default whenever a safe capability exists — not only for “hard” work. Parallelism changes scheduling only; synthesis, integration, dispositions, acceptance, cleanup, and user decisions stay with the parent. Material ambiguity and complexity-increasing review remedies escalate to the user rather than shaky assumptions. No catalog skill is a hard runtime dependency of another. Worktree isolate/integrate/remove policy is canonical in use-subagents and mandatory after lanes finish; harness launchers (e.g. use-pi-subagents) own only their runtime-state cleanup.

Manual-only skills

These skills are never intended for automatic selection:

• explain — structured Markdown/HTML explanations with optional browser open
• vps-setup-hardening — production-impacting remote work with human authentication, independent access/exposure tests, provider firewall changes, access closure, and reboot approval gates

Invoke them explicitly after installation:

• Pi: /skill:explain or /skill:vps-setup-hardening
• Claude Code: /explain or /vps-setup-hardening
• Codex: $explain or $vps-setup-hardening

Pi and Claude Code honor disable-model-invocation: true; Codex honors the bundled policy.allow_implicit_invocation: false. Other agents may ignore invocation-control metadata, so strict manual-only prevention is not claimed for them. Each skill also contains a defensive in-body stop rule.

Runtime and related skills

• awesome-tests works standalone and supports soft co-activation. Use it directly for scoped test planning, authoring, repair, or review. It can co-activate with create-plan when a behavior-changing plan must specify tests or validation, and with code-review when changes materially affect tests or validation. The owning workflow retains artifact, finding, severity, matrix, report, and verdict authority. This is routing behavior, not a hard runtime dependency; all three skills remain independently useful.

• decomplex is a soft integration. It can provide focused advisory reports to code-review, create-plan, and implement-plan when installed and proportionate. Decision-related reviews default to adrs/work/-decomplex.md, while explicit no-write requests return findings in chat; it never edits reviewed targets. Each owning workflow retains its concise built-in gate and records an honest fallback when the skill or report write is unavailable.

• create-report is for explicit standalone HTML reports when no domain-specific workflow owns the task. It is not an automatic completion gate.

• create-changes-report is an optional repository-change handoff. It produces a verified standalone HTML report with expandable critical-code evidence and complete paths when explicitly requested or otherwise selected; implement-plan does not require it for completion.

• use-worktrees owns direct worktree operations. For delegated work, apply it together with use-subagents; the parent agent remains responsible for worktree creation, integration, and cleanup.

• use-subagents is portable policy (delegate-by-default, roles/assignment contract, worktrees/Git/cleanup) for any harness’s built-in tools, plugins, or CLIs. It does not depend on Pi.

• use-pi-subagents is a Pi launcher only — use it with use-subagents when native subagent_* tools are inactive. Never drive competing launchers for the same lane. Parent owns worktrees, Git, and workspace cleanup; Pi clean retires run state only. No unaccounted workflow-owned resources.

• agent-browser remains external. Install it from skills.sh when browser interaction or UI verification is needed:

  npx skills add vercel-labs/agent-browser@agent-browser
  

What each skill does

awesome-tests

Plans, writes, improves, and reviews behavior-focused automated tests and test strategies. It maps tests to observable behavior and material risk, selects the narrowest credible repository-conventional layer, checks failure sensitivity, determinism, and isolation, and assesses flaky, brittle, misleading, redundant, skipped, or false-green tests contextually rather than treating counts or coverage as quality gates.

code-review

Evidence-bound generic and plan-backed reviews. Delegates read-only lanes by default; admits only material reachable findings; parent consolidates scores/verdicts. Relevant accepted ADRs constrain implementation, and decision-related reports default to adrs/work/. Supports focused closure rounds without reopening broad scope.

create-changes-report

Creates a self-contained interactive HTML report after repository changes. The report provides a five-minute overview with drill-down evidence, architecture flow, verification commands and results, expandable critical-code evidence with complete paths, calibrated findings, decisions, risks, and review guidance; mandatory light, dark, and popover browser checks verify the artifact before handoff.

decomplex

Reviews proposed or existing source, plans, architecture, tests, configuration, dependencies, and review recommendations for evidenced unnecessary complexity. It supports Prevention, Audit, and Finding triage modes; writes one advisory adrs/work/ report by default or returns chat-only findings when requested; and never edits reviewed targets.

explain

Manual-only. Produces concise, context-grounded Markdown plus polished standalone HTML for code, systems, and prior plans, reviews, or agent decisions. Explanations can use tables, syntax-highlighted code, and an optional Mermaid diagram. The Node 20+ renderer opens completed HTML in the default browser by default; pass --no-open for CI, tests, SSH/headless/remote work, or other contexts where browser side effects are inappropriate. After a one-time npm ci --ignore-scripts --no-audit --no-fund from the installed skill directory, generated HTML is standalone/offline and needs no server or network access.

create-plan

Research → smallest plan → review → deliver. Delegates research/review by default; asks the user instead of shaky assumptions; maintains one adrs/work/.md document for research, tasks, evidence, review closure, and handoff, with separate ADRs only for significant decisions.

create-report

Creates concise, evidence-backed, actionable standalone HTML reports for decisions, assessments, research, and status outcomes when the user asks for a polished HTML report and no domain-specific workflow owns the task. It is not an automatic completion gate. The executive layer stays about five minutes; headline numbers and key conclusions drill down to labelled evidence; findings and recommendations stay honest about unknowns.

create-skill

Creates, rewrites, reviews, and evidence-backed improves Agent Skills with literal activation routing, concise execution contracts, progressive disclosure, exact resources, trigger/near-miss evaluation, and representative output checks. It includes a placeholder-safe starter and quality checklist.

create-slides

Creates and materially redesigns polished, dependency-free HTML slide decks (vanilla HTML/CSS/JS) from five templates — dark-marker, light-editorial, midnight-tech, bold-keynote, minimal-mono — that share one archetype vocabulary, five named reveal presets, a cross-slide title morph, and runtime plus composition QA. Art-direction intake asks for the look, delivery mode, reveal model and density. Decks open from a local index.html with no server, build step, or network access; PDF and MP4 export (1080p/2K/4K, reveals optionally timed to a caption track) need Node, Chrome and ffmpeg.

create-wiki

Creates and updates reusable technology knowledge bases under wiki//. It researches authoritative, version-matched sources and writes focused files covering mental models, workflows, examples, pitfalls, diagnostics, and trade-offs. Wiki content remains portable and excludes project documentation, architecture decisions, plans, and index/README files.

generate-image

Generates AI images through fal.ai via a small Bun HTTP/queue CLI. Defaults to openai/gpt-image-2, uses FAL_KEY, and downloads result files locally for inspection.

implement-plan

Resumes the shared adrs/work/.md document, reconciles its tasks and accepted ADRs, then runs a delegated loop per item: analyze → implement → check → review → disposition → update → cleanup. The parent owns shared work, ADR status, integration, acceptance, and cleanup. Completion requires verified or approved-descoped tasks, final validation, a clear plan-backed review, and no material open work; HTML reports are optional.

use-worktrees

Creates isolated worktrees under ~/worktrees//, keeps their branches synchronized from the target branch, and runs repository checks after merges. It merges back only on explicit user request, verifies the final diff and push, and removes the worktree and branch only after successful integration.

use-subagents

Harness-agnostic subagent playbook: delegate-by-default, scout/research/worker profiles, assignment contract, worktree isolate/integrate/remove, mandatory cleanup, supervision, and parent verification. Works with whatever launcher the host provides. No dependency on use-pi-subagents.

use-pi-subagents

Pi RPC launcher (scripts/subagents.mjs) when native subagent_* tools are inactive. Complements use-subagents (policy) rather than replacing it. Parent supplies cwd/worktrees, verifies, integrates, removes safe workspaces per policy, then runs script clean for Pi run state only. Retains and reports unsafe/unknown resources.

use-mcp

Uses mcporter as a narrow adapter for configured or ad-hoc MCP servers. It discovers only the requested server and tool, prefers structured calls and cached credentials, treats MCP definitions/results as untrusted, requires approval before OAuth persistence or unfamiliar stdio execution, and documents mcporter’s plaintext file-backed credential storage. Requires mcporter 0.13.3+ on PATH.

web-research

Researches current web and external technical information with available search, retrieval, repository, document, media, and browser capabilities. It favors official, version-matched evidence; small direct lookups can remain artifact-free, while substantive or conflicting research is retained in the active adrs/work/ document and reusable technology knowledge belongs in the wiki.

vps-setup-hardening

Turns a fresh Ubuntu 24.04/26.04 or Amazon Linux 2023 EC2 host into a verified baseline through inspection-first changes, named key-only administration, optional Tailscale, UFW or provider/Security Group ingress, optional Docker and Node.js, independent exposure tests, reboot approval, and a concise evidence-qualified report. It is manual-only and preserves the current access path until a replacement is proven.

Structure

Each skill is self-contained under skills// with a SKILL.md file and optional references/, assets/, and scripts/ resources.

Validate

for d in skills/*; do
  [ -d "$d" ] && npx -y skills-ref validate "$d"
done

skills-ref 0.1.5 reports disable-model-invocation as an unexpected field even though current Pi and Claude Code support it. For manual-only skills (explain, vps-setup-hardening), retain that required safety field and treat only that exact diagnostic as a known external-validator limitation; all other diagnostics remain failures.

Validate catalog metadata and local Markdown links:

node scripts/validate-skill-metadata.mjs skills
node scripts/validate-skill-links.mjs README.md skills

Verify CLI discovery without telemetry:

DISABLE_TELEMETRY=1 npx -y skills@latest add . --list

License

MIT