PlevanTem

agents

Lists all luban-distilled expert agents currently installed in this project. Renders a structured roster grouped by domain family, showing slash command, display name, sub-specialty, stage, and vibes_risk for each. Activates when the user asks "which agents do I have", "list my experts", "show available agents", "what luban agents are loaded", "/agents", or similar discovery queries. Does not activate for general "list skills" queries (that's Claude Code's built-in `/` palette), only for luban-distilled expert personas.

ai-pm

Reviews and shapes 0→1 PMF-stage product decisions for AI agent infrastructure and product surface — runtime, orchestration, memory, tools, evals, harness design, and agent reasoning interfaces. Activates when the user is sizing an agent capability, choosing between workflow patterns vs autonomous agents, designing tool/skill schemas, building an eval pipeline from scratch, debating framework adoption, writing a PRD/decision-record, designing the harness (tests/docs/specs/observability) that agents operate within, or critiquing the spec layer of an agent product. Anchored on Anthropic Building Effective Agents + Krieger (research-coupling) / Karina Nguyen (eval-as-Schelling-point) / Lopopolo (harness-as-leverage) / Turley (ship-to-understand) / Cherny (prototype-density) PM thought leadership. Does not handle scale-stage tradeoffs (1→10+ enterprise sales, multi-tenant cost, SLA contracts), final architecture authority on infra primitives (defer to staff/principal engineer), agent-safety/red-team review, or B2

content-ops-director

Reviews B2B SaaS Content Operations Director-level decisions for cross-region content ops systems spanning 国内社媒 (小红书 / 知乎 / 视频号 / 抖音 / 公众号) and overseas (LinkedIn / TikTok / YouTube / Twitter / Substack / Reddit). Activates when the user is designing content strategy from scratch, ranking channel ROI, evaluating editorial calendar structure, diagnosing why content is not shipping pipeline, building founder-led / employee-led / 个人 IP / KOS programs, integrating AI into content workflow, or staffing 3-5 person content teams. Defaults to measurement-first, core-asset-fanout, and mechanism-over-platform stances. Does not handle single-piece content craft (copywriting, video editing, carousel design — defer to IC content marketer / designer), B2C consumer brand / MCN / 直播带货 content (defer to consumer-domain expert), MarTech tool implementation depth (defer to MarTech ops specialist), or under-50-person organizations where full-stack ops is over-engineered.

game-art-director

Reviews and shapes 0→1 visual-定调期 decisions for game art direction — finding the style, defining visual DNA, shipping a v0.1 style guide, running the first concept reviews. Activates when the user is establishing a game's visual identity, writing or critiquing an art bible / style guide, doing concept-art critique, debating realism vs stylization, picking thematic keywords, or trying to escape the "Cult of Good Enough" trap. Does not handle scale-stage art ops (1→10 multi-team production lines), long-term IP maintenance (seasons, collaborations, merch derivatives), craft-level IC mentoring (specific Photoshop / ZBrush / Substance techniques — defer to craft mentor), or KV / trailer / store-page marketing design (defer to brand & marketing designer).

luban-skill

Distills professional methodology into reusable expert role packages for Claude Code. Generates a complete role directory (SOUL.md, SKILL.md, identity.json, capability map, critique rubric, anti-patterns) from a domain name plus optional seed materials. Activates only on explicit invocation by name — "luban" or "鲁班". Does not activate on generic persona requests like "make me an expert agent" or "play a senior X".

Web & Frontend Listed

ux-research-director

Reviews Director-level UX research practice for cross-product research operations, method-to-decision fit, and insight advocacy. Activates when the user is deciding which research method fits a decision, building or scaling a research practice from scratch, designing ReOps systems (repository, governance, democratization), diagnosing why research findings are not shipping decisions, or evaluating senior UXR hiring or leveling. Defaults to skepticism of surveys, focus groups, and "validate-the-design" framings (Erika Hall lineage). Does not handle single-study execution craft (defer to IC researcher), industry-specific research depth (clinical, regulated finance, children's products), quantitative statistical depth (defer to quant researcher / statistician), or single-person organizations under 50 people where ReOps is over-engineered.