River-181/harness-engineering-skills

agent-loop

Use when designing AI Agent product capabilities, competition judging evidence, or production agent safety. Creates judgment → action → verification/improvement loops with retrieval, rule checks, approval, logging, and feedback.

原文语言：英语

architecture

Use after PRD/domain model and before implementation. Designs system architecture, component boundaries, dataflow, integration points, security boundaries, logging, audit, failure handling, and demo-vs-production separation.

原文语言：英语

code-rules

Use before asking coding agents to implement features, especially when outputs must converge across Claude, Codex, Gemini, or human maintainers. Creates deterministic code-generation rules for naming, imports, exports, folders, APIs, UI states, tests, lint…

原文语言：英语

cps-framing

Use when the team needs a clear why, strategic narrative, problem definition, or scope boundary before PRD or implementation. Frames a product or feature using Context → Problem → Solution.

原文语言：英语

ddbm-business-model

Use before CPS, PRD, pitch, or competition submission when a product needs a sharp data-driven business model. Creates the 11-block DDBM canvas, data asset assumptions, risk linkage, cost/revenue model, and judge-facing business thesis.

原文语言：英语

demo-handoff

Use before presentation, judging, client delivery, sprint closeout, or team handoff. Prepares final demo script, runbook, handoff document, known limitations, change history, maintainer notes, and submission-readiness package.

原文语言：英语

domain-model

Use before architecture, database design, code generation, or AI Agent implementation. Builds an implementation-ready domain model with entities, relationships, states, transitions, events, permissions, and audit relationships.

原文语言：英语

eval-design

Use when a project needs proof that agent outputs remain useful, grounded, safe, and maintainable over time. Creates organization-specific evals: golden cases, rubrics, failure modes, regression schedule, cross-agent normal-form comparison, and safety gates.

原文语言：英语

init-project

Use at the start of a product, AI Agent, MVP, or enterprise delivery repo, or when a repo lacks harness.yaml, docs, rules, evals, and CLAUDE.md project memory. Creates a complete local harness scaffold.

原文语言：英语

intake-log

Use before CPS, PRD, or implementation when raw transcripts, notes, client conversations, brainstorms, or vague requirements need normalization. Converts them into source logs, meeting logs, explicit/implied requirements, decisions, open questions, and build…

原文语言：英语

jb-finai-adapter

Use for JB금융그룹 Fin:AI 본선/예선 work or similar financial AI Agent competitions. Aligns a financial AI Agent MVP, proposal, feature specification, demo flow, change log, and risk controls to judging criteria.

原文语言：英语

prd-spec

Use when moving from framing to buildable scope. Creates PRD, MVP proposal, and implementation-ready feature specs traced to CPS, users, scenarios, acceptance criteria, demoability, and risk controls.

原文语言：英语

principles-definitions

Use when terms drift, the team needs non-negotiables, or downstream docs/code need consistent names. Defines enforceable project principles, canonical vocabulary, anti-synonyms, and concept hierarchy.

原文语言：英语

risk-compliance

Use for regulated, financial, customer-facing, enterprise, or high-risk workflows that need reviewable engineering controls. Creates operational risk, privacy, security, compliance, approval, audit, copyright, hallucination, explainability, and responsibility…

原文语言：英语

route

Use when the user asks what to do next, gives messy product/enterprise AI/FDE/MVP/agent-delivery requirements, asks for a harness, or requests implementation before discovery artifacts exist. Routes the request into the right Harness Engineering lifecycle…

原文语言：英语

skillpack-maintainer

Use when maintaining or releasing the Harness Engineering Skill Pack itself. Adds skills, tunes trigger descriptions, shrinks or expands playbooks, adds templates/evals, updates scripts, runs audits, and prepares release notes.

原文语言：英语

surface-map

Use when a vague product may require multiple apps, consoles, portals, dashboards, APIs, actors, permissions, or action boundaries. Decomposes a product into user-facing and operator-facing software surfaces by actor, job, permission, data access, action…

原文语言：英语

validate-harness

Use before implementation, submission, demo, handoff, or public release. Validates the current project harness for required files, sections, placeholders, traceability, gates, risk disclosures, and readiness.

原文语言：英语