菜单
USE FOR: Verifying technical claims, comparing options, and investigating unfamiliar libraries/APIs/tools with cited, confidence-rated evidence. DO NOT USE FOR: implementation, opinion without evidence, or tasks better served by a narrower repo-specific skill (e.g. `data-platform`, `collaboration`).
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
| name | technical-research |
| license | MIT |
| metadata | {"version":"1.0.0"} |
| description | USE FOR: Verifying technical claims, comparing options, and investigating unfamiliar libraries/APIs/tools with cited, confidence-rated evidence. DO NOT USE FOR: implementation, opinion without evidence, or tasks better served by a narrower repo-specific skill (e.g. `data-platform`, `collaboration`). |
Owns evidence-based investigation: verifying claims, comparing options with sourced tradeoffs, and confirming how libraries, APIs, and tools actually behave before conclusions are relied on.
Clarify scope: what exactly needs to be answered?
Search official docs first (WebSearch + WebFetch)
Check GitHub repos: issues, discussions, source code for implementation details
# Clone to /tmp for analysis when needed
git clone --depth 1 <repo-url> /tmp/<repo-name>
Cross-reference: do multiple sources agree? Flag contradictions.
Test when possible: run code, check versions, verify claims hands-on
When synthesis needs categories, labels, or first-use terms before final
wording, use clarify-concepts after evidence acquisition and before
drafting conclusions.
USE FOR: This dotfiles environment and its agent harness: machine setup (Ubuntu, WSL2, macOS, home-manager, nix-darwin), daily Nix operations and upgrade/recovery, missing CLI tools / command not found, operating concepts, clipboard (Vim, Neovim, tmux), Claude Code and Codex CLI config, hooks, MCP, skill installation, postman routing, orchestrator runbooks, tmux workspaces, issue/PR worktrees, pane operations, prompt/review contracts, resume handoff; Agent Skills authoring: add/edit/validate skills, Waza, publishing. DO NOT USE FOR: Nix package authoring (programming).
USE FOR: Guardian/critic native multi-perspective review workflow, peer critic evidence, material-finding convergence, and blocking-finding synthesis. DO NOT USE FOR: implementation, dispatcher fan-out, overrides, or public posting.
USE FOR: Clarify concepts before names harden by fixing concrete referents, roles, sequence, candidate terms, and first-use definitions. Use for terminology-heavy design docs, research synthesis, cause-isolation or mitigation plans, naming decisions, and reasoning-order summaries only when concept/referent ambiguity could distort the work. DO NOT USE FOR: ordinary terminology/glossary editing when referents are stable, simple editing, translation-only work, proofreading, generic summarization/research, implementation, or broad content generation.
USE FOR: Implementation plans and durable task tracking: reduce ambiguity, compare options, parallel investigation, multi-source synthesis, review gates, task artifacts, evidence logs, handoff/resume, and DONE/BLOCKED verification. DO NOT USE FOR: unrelated tasks, broad rewrites outside the request, or generated runtime outputs.
USE FOR: Prose QA, Vale/Harper, terminology, English/Japanese edits, English-to-Japanese, stable glossaries, rhythm, AI-slop cleanup, publication QA. DO NOT USE FOR: AI detector/humanizer workflows, fresh generation, meaning-changing rewrites, replacing author judgment, publishing, model/cost/privacy, external sharing, or chunking.
USE FOR: Reference-only external provider packs and dormant source inventory: dbt, azure, databricks-official, obsidian, hashicorp-terraform-codegen, hashicorp-terraform-module, hashicorp-terraform-provider, google, aws, Databricks, Azure, Microsoft Foundry, Google Cloud, GCP, AWS, Terraform, HashiCorp, Obsidian, provider-pack lookup, and promotion questions. DO NOT USE FOR: repo-local guardrails, harness implementation, or making provider packs active without explicit promotion.