Each repo now has an adopt, recreate, install, or fenced outcome.
The Short Version
This was not a download-everything pass. It became a security-reviewed harvest: install only what cleared review, recreate useful behavior when that was safer, and update existing local skills before adding new ones.
The concrete gains are visible now: new debugging and skill-authoring lanes, local UI guidance search, a vibe-check discovery skill, a guarded Lavish AXI wrapper, OpenSpec-style change packets, Job Application Compass scanner/PDF/manual-apply tools, and new Shareables recipes for codebase understanding and visual QA review packets.
Review Scope
The pass covered all repos named across the conversation, including the two later additions and Lavish AXI.
uipro-cli, OpenSpec, and Lavish AXI were inspected in bounded lanes; only safe help/version/offline flows were run.
Public ATS scan, ATS-safe PDF preparation, and manual apply assistant.
Codebase understanding and visual QA review packets, plus Lavish review-loop guidance.
systematic-debugging, skill-authoring, ui-ux-pro-max, vibe-check, lavish-axi, and current-social-research.
Those stayed out after review; the safe local equivalent prepares evidence and copy for human confirmation.
What Actually Changed
Installed after review
ui-ux-pro-max is available as a local searchable design guidance skill. vibe-check is installed as a static local skill. OpenSpec and Lavish AXI CLIs were installed only in review workspaces for bounded inspection.
Recreated locally
Career Ops became safe Job Application Compass tools. Understand Anything became a Shareables codebase-understanding scanner/report lane. Last30Days became a current-social-research planning script and Shareables freshness gate.
Updated existing skills first
frontend-ux, design-taste-gate, PRD strategist, project-structure-architect, plain-to-dev, gh-fix-ci, gh-address-comments, Playwright, and GitHub conventions all absorbed relevant patterns.
Added only real gap skills
systematic-debugging and skill-authoring cover recurring gaps that did not have clean local owners. lavish-axi and vibe-check became dedicated skills because they own distinct workflows.
Strengthened MCP workflows
Shareables gained codebase-understanding, visual QA review packet, and Lavish pilot guidance. Job Application Compass gained application-operations tools with tests.
Kept risky automation fenced
Browser/cookie flows, hooks, daemons, auto-submit, auto-send, private scraping, and telemetry-heavy defaults were not made part of the default local system.
Install Versus Recreate Versus Pattern
| Decision | Repos | Local result |
|---|---|---|
| Install as local skill | TexasBedouin/vibe-check; nextlevelbuilder/ui-ux-pro-max-skill | Static/local skill surfaces are available and linked from planning/design skills. |
| Install only in review workspace | Fission-AI/OpenSpec; kunchenguid/lavish-axi | Version/help/playbook checks ran with telemetry disabled; no project mutation, server, poll, or dashboard start. |
| Recreate safe behavior | santifer/career-ops; Egonex-AI/Understand-Anything; mvanhorn/last30days-skill | JAC tools, Shareables scanners/reports, and current-social planning now exist locally. |
| Harvest patterns into existing skills | obra/superpowers; anthropics/skills; mattpocock/skills; garrytan/gstack; addyosmani/agent-skills; Leonxlnx/taste-skill | Debugging, skill authoring, product/design review, QA, taste, and planning skills were updated. |
| Fence runtime-heavy automation | GStack hooks/browser surfaces; Career Ops auto-submit/send; Last30Days cookies/APIs; Lavish server/poll; Understand Anything dashboard/plugin | Useful ideas adopted, but no persistent browser, daemon, private platform scraping, auto-send, or auto-submit default was added. |
Repo-by-Repo Takeaways
Open each row for the practical takeaway, the local impact, the gap it filled, and the security/install outcome.
obra/superpowers - systematic debugging and verification discipline
Takeaway: the strongest reusable part was not the hook system. It was the insistence on root-cause evidence, test-first pressure, stop conditions, and verification before declaring victory.
Local impact: created the systematic-debugging skill and linked it into CI repair, PR comment response, GitHub conventions, and Playwright-supported bug work.
Gap filled: Codex now has a named local lane for reproduce, localize, hypothesize, fix, regression-test, and recheck instead of patching by feel.
Security/install outcome: MIT source reviewed. SessionStart hooks and behavior-injection setup stayed out; the pattern was recreated locally.
- Repo: obra/superpowers
- Local outcome: new systematic-debugging skill plus existing-skill links.
anthropics/skills - skill authoring shape and progressive disclosure
Takeaway: useful skills stay small at the entrypoint, push long detail into references, reuse scripts/assets, and expose enough trigger guidance for an agent to choose them correctly.
Local impact: created the skill-authoring skill and strengthened the update-existing-first rule for future harvests.
Gap filled: the local system now has a repeatable way to decide whether external material belongs as a patch, helper script, reference, or new skill.
Security/install outcome: no root license was found in the clone, so this was used as pattern reference only, not vendored wholesale.
- Repo: anthropics/skills
- Local outcome: new skill-authoring skill and skill-harvest rubric.
mattpocock/skills - codebase design, issue shaping, and teaching posture
Takeaway: the useful pattern was turning ambiguity into domain language, ownership boundaries, tests, and understandable explanations.
Local impact: project-structure-architect gained deeper module quality checks; systematic-debugging absorbed practical bug-investigation posture.
Gap filled: local planning can now ask whether an interface is clear, local, testable, removable, and useful before introducing a new abstraction.
Security/install outcome: MIT static skill repo. No runtime install needed.
- Repo: mattpocock/skills
- Local outcome: project-structure and debugging updates.
garrytan/gstack - product review, QA, and office-hours patterns without the daemon risk
Takeaway: the best parts were review checklists, investigation routines, office-hours forcing questions, and design/product critique structure.
Local impact: PRD strategist, project-structure-architect, design-taste-gate, frontend-ux, and Playwright QA now carry more explicit review and ship-readiness language.
Gap filled: we gained stronger design/product review posture without importing a background agent stack.
Security/install outcome: MIT source reviewed. Setup scripts, browser/cookie surfaces, hooks, telemetry-like workflow, and daemon behavior stayed fenced. Patterns were recreated in local skills.
- Repo: garrytan/gstack
- Local outcome: recreated guidance only, no default runtime.
nextlevelbuilder/ui-ux-pro-max-skill - searchable design guidance installed
Takeaway: the clear value was a searchable design-system and UX guidance database that can be queried before frontend work or design critique.
Local impact: installed the uipro CLI after package review, initialized it offline for Codex, created a local ui-ux-pro-max skill, and linked it from frontend-ux and design-taste-gate.
Gap filled: when no house design system is supplied, Codex now has a local design lookup instead of improvising style direction.
Security/install outcome: MIT package reviewed; npm metadata showed no install scripts. Install used ignore-scripts first, then offline init. No paid service required.
- Repo: nextlevelbuilder/ui-ux-pro-max-skill
- Local outcome: installed local ui-ux-pro-max skill and verified search script.
Egonex-AI/Understand-Anything - static codebase understanding lane for Shareables
Takeaway: codebase explanation should start with entrypoints, domain terms, data flow, dependency risks, and progressive tour paths before asking for a live dashboard.
Local impact: Shareables MCP gained a codebase-understanding report recipe, a bounded scan script, workflow docs, quality gates, and project-map entries.
Gap filled: Codex can now create a public-safe codebase onboarding Shareable without installing a dashboard/plugin runtime.
Security/install outcome: MIT source reviewed. Plugin/dashboard runtime stayed out; the safe static scanner/report path was recreated locally.
- Repo: Egonex-AI/Understand-Anything
- Local outcome: Shareables codebase-understanding report lane.
addyosmani/agent-skills - production agent discipline and review loops
Takeaway: useful agent skills make source grounding, implementation quality, verification, and anti-rationalization checks explicit.
Local impact: skill-authoring, systematic-debugging, gh-fix-ci, gh-address-comments, and Shareables quality guidance now carry more explicit evidence and verification requirements.
Gap filled: the system has clearer expectations for when an agent is allowed to say done.
Security/install outcome: MIT static skill repo. No duplicate command suite was installed.
- Repo: addyosmani/agent-skills
- Local outcome: adopted into local agent discipline and review skills.
santifer/career-ops - scanner, PDF export, and apply assistance recreated in Job Application Compass
Takeaway: the useful pieces were public ATS source scanning, local ATS-safe PDF preparation, application-mode question extraction, and packet-grounded answer drafting.
Local impact: Job Application Compass gained scan_public_job_sources, prepare_ats_pdf_export, and prepare_manual_apply_assistant, with MCP surface tests and self-test coverage.
Gap filled: JAC now has concrete application-operations tools instead of only planning notes.
Security/install outcome: MIT source reviewed. We recreated the safe scanner/PDF/manual-apply core. Private scraping, autofill, auto-submit, CRM, and automatic sending remain blocked because they create platform, consent, reputation, and browser-session risk.
- Repo: santifer/career-ops
- Local outcome: three new JAC MCP tools, no auto-submit or auto-send.
Leonxlnx/taste-skill - specific taste language for UX and visual QA
Takeaway: taste improves when the agent can name the intended visual register, contrast decisions, density, trust cues, motion, and what makes the result specific rather than generic.
Local impact: frontend-ux and design-taste-gate gained design-control dials and more concrete critique language.
Gap filled: visual critique is less likely to collapse into vague polish words.
Security/install outcome: MIT source reviewed. No installer script was run; patterns were folded into local UX/design skills.
- Repo: Leonxlnx/taste-skill
- Local outcome: design taste and frontend UX skill updates.
mvanhorn/last30days-skill - current social research without cookie/API sprawl
Takeaway: recent public/community signal research is a real gap, but it needs source boundaries, citation rules, uncertainty labels, and explicit excluded-source notes.
Local impact: current-social-research gained a planner script and adoption notes; Shareables gained current-research gates for freshness, exclusions, and caveats.
Gap filled: Codex can now plan a 30-day-ish public signal pass without pulling in cookies, private APIs, webhooks, or a hosted runtime.
Security/install outcome: MIT source reviewed. We recreated the safe research-planning lane; no X cookies, private API keys, webhooks, or MCP runtime were installed.
- Repo: mvanhorn/last30days-skill
- Local outcome: current-social-research script and Shareables current-source gates.
Fission-AI/OpenSpec - change packets plus isolated CLI review
Takeaway: implementation-affecting work benefits from proposal, spec, design notes, tasks, verification, and archive records.
Local impact: PRD strategist and project-structure-architect gained an OpenSpec-inspired create_change_packet.py helper and adoption guidance.
Gap filled: planning now has a lightweight on-disk change lifecycle without requiring a project to adopt OpenSpec wholesale.
Security/install outcome: MIT npm package reviewed. The CLI was installed only in the review workspace with telemetry disabled for version/help checks. No project init, update, dashboard, or global install was run.
- Repo: Fission-AI/OpenSpec
- Local outcome: change-packet helper and isolated reviewed CLI.
TexasBedouin/vibe-check - beginner build discovery installed as a local skill
Takeaway: early app work needs a humane translation layer: problem, user, outcome, constraints, done criteria, opportunity map, story map, and buildable plan.
Local impact: installed vibe-check as a local static skill, linked it from plain-to-dev and PRD strategist, and kept its diagrams/blueprint references available.
Gap filled: Codex now has a dedicated path for beginner app discovery and visual blueprinting instead of turning a vague idea into immediate implementation.
Security/install outcome: MIT source reviewed. Static skill copied locally; release helper scripts were not run. Version check remains optional and non-blocking.
- Repo: TexasBedouin/vibe-check
- Local outcome: installed local vibe-check skill plus plain-to-dev integration.
kunchenguid/lavish-axi - guarded HTML artifact review loop for Shareables
Takeaway: artifact review gets stronger when browser-detected layout issues and human annotations can be returned as structured comments.
Local impact: created the lavish-axi skill, added a Shareables build-time review-loop option, and documented how to keep pilots sandboxed and non-exporting.
Gap filled: Shareables now has a named way to pilot HTML annotation/layout-gate loops without making an external server/poll process the default.
Security/install outcome: MIT npm package reviewed. Installed only in the review workspace, telemetry disabled for help/playbook checks. No server, poll loop, browser session, or project write was started.
- Repo: kunchenguid/lavish-axi
- Local outcome: local lavish-axi skill and Shareables pilot guidance.
Capability Map
| Local surface | Added or changed capability | Source influence |
|---|---|---|
| Shareables MCP | codebase-understanding report recipe and bounded scanner | Understand Anything |
| Shareables MCP | visual-qa-review-packet recipe and screenshot-evidence guidance | GStack, Taste, UI Pro Max, Addy agent skills |
| Shareables MCP | Lavish AXI build-time annotation/layout-gate pilot guidance | Lavish AXI |
| Job Application Compass | public ATS source scanner | career-ops |
| Job Application Compass | ATS-safe PDF export preparation | career-ops |
| Job Application Compass | manual apply assistant with no autofill/submit | career-ops |
| frontend-ux | style-source, design-dial, and anti-default UI guidance | UI Pro Max, Taste, Anthropic frontend-design, GStack |
| design-taste-gate | specific taste critique, visual QA, and search-before-judgment path | Taste, UI Pro Max, GStack |
| prd-strategist | OpenSpec-style change packets and vibe-check beginner translation | OpenSpec, vibe-check, GStack office hours |
| project-structure-architect | module quality lens and codebase tour map | Matt Pocock skills, Understand Anything, OpenSpec |
| gh-fix-ci and gh-address-comments | systematic reproduction and classification before edits | Superpowers, Addy agent skills, Matt debugging |
| current-social-research | 30-day public signal planning with citations and exclusions | last30days |
| plain-to-dev | beginner build translation and scope checkups | vibe-check |
Technical Verification
This section is intentionally dry. It records what was actually checked locally.
Shareables MCP tests
- python3 -m py_compile shareables_core.py passed.
- Targeted pytest for artifact routing/design-system/schema checks passed: 3 passed.
- visual-qa-review-packet route assertion added.
Job Application Compass tests
- python3 -m unittest tests.test_mcp_surface -v passed: 15 tests OK.
- python3 server.py --self-test passed and reported 29 tools.
- New tools are visible in the MCP surface contract.
Local helper scripts
- current_social_plan.py compiled and produced JSON through python3 -m json.tool.
- create_change_packet.py generated proposal, spec, design, tasks, verification, and metadata.
- ui-ux-pro-max search returned local design-system guidance.
Reviewed install surfaces
- uipro-cli package metadata showed no install/postinstall scripts before offline init.
- OpenSpec was installed in a review workspace with ignore-scripts and telemetry disabled for version check.
- Lavish AXI was installed in a review workspace with telemetry disabled for help/playbook checks only.
- vibe-check was copied as a static local skill; release/bump scripts were not run.
Browser and automation boundaries
- No GStack setup/hooks/daemon/browser-cookie workflow was run.
- No Career Ops auto-submit or automatic sending workflow was enabled.
- No Last30Days cookies, API keys, webhooks, or hosted runtime were installed.
- No Lavish server, poll loop, browser session, or project mutation was started.
Public Guardrails
The stronger pass was less conservative about useful integrations, but it still kept trust boundaries where the security review found runtime, platform, or consent risk.
What is now allowed
- Use ui-ux-pro-max as a local searchable design guidance source when no design system is supplied.
- Use vibe-check for beginner app discovery, visual blueprinting, and scope translation.
- Use Job Application Compass public scan, PDF preparation, and manual apply tools for application packets.
- Use Shareables codebase-understanding and visual QA review packet recipes for public artifacts.
- Use Lavish AXI only through the guarded local skill for sandboxed HTML review pilots.
What still needs explicit separate review
- Do not run external setup scripts, hooks, daemons, browser profile automation, or cookie workflows by default.
- Do not auto-submit job applications or automatically send outreach from Job Application Compass.
- Do not scrape private or authenticated platforms without explicit platform-policy and security review.
- Do not publish private local paths, secrets, draft notes, or browser/session artifacts in public Shareables.
Original Repos
- obra/superpowers: Systematic workflows, hooks, and verification patterns.
- anthropics/skills: Skill structure, references, scripts, and progressive disclosure examples.
- mattpocock/skills: Debugging, TypeScript/codebase design, and teaching-oriented agent skills.
- garrytan/gstack: Review, QA, design, investigation, and runtime-heavy agent stack patterns.
- nextlevelbuilder/ui-ux-pro-max-skill: Design system, UX, style, and frontend guidance database.
- Egonex-AI/Understand-Anything: Codebase understanding and dashboard/plugin inspiration.
- addyosmani/agent-skills: Production agent skills, source grounding, and implementation discipline.
- santifer/career-ops: Job application operations patterns.
- Leonxlnx/taste-skill: Taste and anti-generic design critique patterns.
- mvanhorn/last30days-skill: Recent public signal and social research pattern.
- Fission-AI/OpenSpec: Change lifecycle and spec-driven planning pattern.
- TexasBedouin/vibe-check: Beginner app discovery, opportunity mapping, and visual blueprinting.
- kunchenguid/lavish-axi: HTML artifact annotation, layout-gate, and review loop inspiration.