Agent Automation & Development Skills

Use for AGENT DX — the surface an AI agent consumes as a developer. DO — keep an in-progress AI/Agent SDK, tool, structured-output, error, or telemetry change minimal and agent-consumable. REVIEW — audit an agent-facing SDK/tool/error/telemetry surface for stable contracts, agent recovery, and trust-boundary safety, then score it. DESIGN — shape a new agent-facing surface (loop, tools, MCP, structured output, error envelope, SDK telemetry) or explain a principle. Triggers on 'design our Agent SDK', 'are our tool schemas agent-safe', 'review our MCP tool surface', 'shape retry-able tool errors for the model', 'is our agent telemetry leaking PII'. Do NOT use for human developer API/SDK/CLI surfaces (use dx-audit / dx-design), agent-readable docs/AGENTS.md/llms.txt (use agent-docs), repo agent-hardening gates/hooks (use harden-repo-for-coding-agents), or operating eval/observability loops (use agent-ops).

Use for AGENT UX — the interaction surface an AI agent acts THROUGH as an end user (a UI, app, or computer-use target), often on a human's behalf. DO — make a surface an agent can perceive, target, and act on safely. REVIEW — audit a surface for machine-readable state, deterministic actions, agent agency/approval, and human-vs-agent conflict, then score it. DESIGN — shape the machine-readable state, the stable action handles, the approval/authority gate, or the human/agent dual path, or explain a principle. Triggers on 'can our agent actually use this UI', 'make our app agent-operable', 'should the agent confirm this action', 'our agent clicks the wrong control', 'expose our app state for computer-use'. Do NOT use for human-only product UX or visual design (use ux-audit / ui-design), docs an agent reads (use agent-docs), the SDK/tool schema or token-exchange mechanism (use agent-dx), or operating the agent loop (use agent-ops).

Executes tasks using the Claude Code CLI (claude). Automatically determines the safest permission mode (plan, acceptEdits, or bypassPermissions) based on the task type.

Use to plan, sharpen, run, or make sense of customer discovery interviews — plan an interview (set a learning goal, choose who to talk to, write a screener, build a recruiting cadence), critique and rewrite questions that lead, pitch, or fish for compliments, coach the live conversation (rapport, silence, probing, asking for a real commitment), or synthesize raw notes into an evidence-backed interview snapshot. Triggers on 'are these good interview questions', 'plan customer/discovery interviews', 'how do I run a user interview', 'who should I talk to', 'how do I make sense of my interviews', 'rewrite this leading question'. Do NOT use to validate a market or decide whether to build something (use research), to usability-test a built interface or flow (use ux-audit), or to write up findings as a polished memo or report (use writing-design).

Use to DESIGN or REVIEW software, docs, SDKs, and repos so AI agents can consume them as a first-class audience — agent experience (AX), the agent-facing analog of UX and DX. Covers agent-readable docs (llms.txt, AGENTS.md, MCP and tool descriptions, RAG-friendly structure), AI and Agent SDK design (streaming, tools, structured output, agent loops, tracing), repo agent-readiness review, and human-versus-agent audience conflicts. The umbrella AX discipline — it routes to harden-repo-for-coding-agents to HARDEN a repo, rules-from-coding-agent-failures to PROMOTE observed failures into rules, and agent-evals to INSTRUMENT an AI product's eval loops. Triggers on 'make our app agent-friendly', 'review our llms.txt for agents', 'design an Agent SDK', 'is this AGENTS.md agent-ready', 'AX review'. Do NOT use for human developer experience (use dx-audit or dx-design), end-user UX (use ux-audit), or to actually scaffold a repo's gates and hooks (use harden-repo-for-coding-agents).

Use to ASSESS, HARDEN, SCAFFOLD, or DIAGNOSE a repository's agent-readiness for coding-agent harnesses such as Claude Code, Cursor, Codex, Copilot, Windsurf, and Aider.

Use to shape or fix a product experience as a story — map an experience as a narrative arc with the user cast as the protagonist (concept, origin, or usage story), diagnose a flat or broken experience by finding the missing beat (no clear goal, no inciting incident, anticlimax, or the gap where users drop out), or craft a concept or origin story to align a team or pitch the experience vision. Triggers on 'map our onboarding as a story', 'our experience feels flat — diagnose the arc', 'where do users drop in the story', 'help me tell the product's concept story', 'lay this flow onto a narrative arc'. Do NOT use to heuristically audit usability or accessibility of a built interface (use ux-audit), to write a narrative as prose, a memo, or a talk (use writing-design), or to decide what to build or frame the opportunity (use product-discovery).

Use to keep code MINIMAL and right-sized for AI coding agents. DO — keep an in-progress change minimal (reuse before adding, subtract, avoid the wrong abstraction, check blast radius). REVIEW — audit existing code or a repo for slop (duplication, dead code, over-engineering) and parallel-readiness (coupling, boundaries, gate coverage), then score it. DESIGN — shape right-sized boundaries and work-partitioning so many agents work in parallel, sequence a safe refactor, or explain a principle. Triggers on 'is this over-engineered', 'reduce the slop', 'audit our coupling/boundaries', 'structure this repo for parallel agents'. Do NOT use to wire enforcement gates/hooks/AGENTS.md (use harden-repo-for-coding-agents), for developer API/SDK/CLI surfaces (use dx-audit / dx-design), for product UX (use ux-audit / ui-design), or to tighten prose (use writing-audit).

Use when you operate coding-agent infrastructure with a feedback signal — eval suites, run-level telemetry, A/B baselines, or a skill catalog under test — and want to…