| name | working-with-ai |
| description | Teams working with AI as a tool: coding agents on real repos, AI-generated UI repair, and org enablement. Use for AGENTS.md/CLAUDE.md rules, vibe-coding security, multi-agent workflows, planning big work as agent-sized decision tickets, layout diagnosis, judge loops, MCP/context archives, or internal agents. AI inside product → ai-experience-design; user-resculptable software → malleable-software. |
Working with AI
Sources/gaps: references/sources.md maps the agentic-coding, AI-UI-direction, and enablement corpus; references/coverage-gaps.md tracks fast-moving tool/workflow candidates.
The boundary that matters: this skill is about you and your team using AI as a tool — agents in your repo, AI output on your screen, AI rollout in your org. If you're designing AI features into your product for end users (confidence, attributions, corrections, agent UX), that's ai-experience-design. If end users are reshaping the product itself with AI, that's malleable-software.
Read exactly one reference file below; each carries the full distillation for its area (sources and extended notes live in a same-named subdirectory next to it).
- Agentic coding — running coding agents on real projects: the constraint stack, AGENTS.md/CLAUDE.md rules, distilling taste into rules, the non-delegable human review role, vibe-coding security holes, worktrees/multi-agent topologies, designer-in-the-repo: references/agentic-coding.md (extended practitioner quotes: references/agentic-coding/ai-era-field-notes.md)
- AI UI direction — an AI produced UI and you have to make it right: diagnosing layout bugs in primitive vocabulary (fixed/fill/hug, padding vs gap, nesting), the named-failure catalog, prompt-vs-edit, the independent-judge loop: references/ai-ui-direction.md
- AI enablement — making one person's AI workflow an org capability: decision archives wired to MCP, company-specific internal agents, promoting personal context to team skills, enablement programs and adoption mechanics: references/ai-enablement.md
Teaching agents a product standard (Vercel pattern): when the request is to make agents follow a team's product/design judgment, do not stop at prompt advice. Build the loop: persistent trigger instructions, a lean skill entry point, routed source references, traceable rule IDs, exemplars from real work, explicit coverage gaps, deterministic checks for mechanical rules, and a human-owned evidence/update review. Use creating-skills for packaging mechanics, design-systems for token/component context, and this skill's agentic-coding/ai-enablement references for repo and rollout mechanics.
Boundary splits to respect:
- AI features in your product →
ai-experience-design — restated because it's the most common misroute: building with AI lives here; building AI into the product lives there.
- Generation quality itself →
frontend-design — de-slop substrate and distinctive aesthetics; the ai-ui-direction reference takes over when output needs direction and repair.
- DS-as-AI-context →
design-systems — wiring a design system into codegen; the ai-enablement reference covers the programs that get people using it.
- Skill-authoring mechanics →
creating-skills — writing the rule/skill files the agentic-coding and ai-enablement references depend on.
