---
name: repl-tech-stack
description: "Generate TECH_STACK.md defining languages, libraries, and implementation constraints. Use when the user asks to create or update TECH_STACK.md. (TECH_STACK.md 작성을 요청받으면 사용)"
---

Generate TECH_STACK.md in two phases.

INPUTS
- PRODUCT_SPEC.md: correct and final. Source of product requirements.
- TECH_DECISIONS: correct and final. Highest-priority source of technology names and versions.
- ARCHITECTURE.md does not exist yet. It will be written AFTER this document, using it as input.

SOURCE RULES
- Final TECH_STACK.md MUST contain only technologies from: (a) TECH_DECISIONS, (b) proposals approved by the user in Phase 1.
- MUST NOT put an unapproved technology in TECH_STACK.md.
- MUST NOT contradict TECH_DECISIONS. If a proposal would conflict with it, drop the proposal.
- MUST NOT restate product behavior from PRODUCT_SPEC.
- If TECH_DECISIONS conflicts with PRODUCT_SPEC, stop and list the conflicts. Nothing else.

LANGUAGE RULES
- Phase 1 output: write in the user's language.
- Conflict lists (SOURCE RULES, Phase 2 edit conflicts): write in the user's language.
- Phase 2 output (TECH_STACK.md): write in English, regardless of the user's language.

PHASE 1: GAP FILL AND PROPOSAL

Step 1. Determine project type from PRODUCT_SPEC and TECH_DECISIONS.
Examples: web service, web app, CLI tool, library/package, desktop app, mobile app, game, batch/data pipeline, browser extension, other.
If the type is unclear, include it as a BLOCKED question.

Step 2. Derive the decision categories this project type needs.
- Always required: runtime/platform + version; language + build/run method; package/dependency manager + lockfile (if the ecosystem has one); test tool; linter; formatter; CI system; distribution/release method and target; license.
- Coverage, only if the project uses coverage: coverage tool + threshold.
- Configuration, only if the project is configurable: config format, setting/env var names, secret storage.
- Type-specific categories: derive them from the project type and the capabilities PRODUCT_SPEC requires (e.g. entry-point framework, argument parsing, persistence, networking/HTTP client, external service clients, auth, background work, UI toolkit, rendering/graphics, packaging/installer, update mechanism).
- Use PRODUCT_SPEC only to decide whether a category applies, never to choose a technology.
- MUST NOT ask about a category the project type does not need.

Step 3. For every category not defined in TECH_DECISIONS, decide it yourself and propose a value. Do NOT ask open questions for gaps you can fill.
- If TECH_DECISIONS does not say whether coverage is used, propose it as a CHOICE (alt: no coverage).

Classify each proposal:
- OBVIOUS: mandated or defaulted by an already-decided runtime, framework, or tool (e.g. its official package manager, lockfile, config format, test runner, linter). Decide it. Do not ask.
- CHOICE: real alternatives exist. Pick one recommended option, name the one main alternative, give a one-line reason.
- BLOCKED: no defensible default without information only the user has (e.g. release target, license intent, project type). Ask one question, with your best guess attached.

Versions: propose a version constraint matching the ecosystem's syntax. Mark any version you are not certain is current with "(verify)".

Phase 1 output rules:
- Add a one-line plain description for every tool.
- Keep the final line verbatim in English (do not translate it).

Phase 1 output format (and NOTHING else):

1. Project type: one line.
2. OBVIOUS list: one line each — item: value.
3. CHOICE list: one line each — item: recommended (alt: X) — reason.
4. BLOCKED list: numbered questions, each with a best-guess default.
5. Final line: "Reply `approve all`, or list changes by item."

Do NOT output TECH_STACK.md in Phase 1.
If there are no gaps, skip Phase 1 and go to Phase 2.

PHASE 2: GENERATE
Run after the user approves or edits the Phase 1 list. Treat the approved list as part of TECH_DECISIONS. Do NOT re-ask approved items. If the user's edits create a conflict, list the conflict and stop.

SCOPE BOUNDARY
- Cover only technology choices and the rules that follow from them.
- MUST NOT define components, layers, modules, boundaries, data flow, responsibilities, or system structure.
- MUST NOT define directory layout beyond paths mandated by the chosen framework or tooling.
- MUST NOT include process rules (change control), product non-goals, or architectural invariants.
- State each rule once. MUST NOT repeat a rule across sections.

OUTPUT FORMAT (Phase 2)
- Language: English only.
- Exactly these 8 sections, in this order, numbered headings, no other sections:
  1. Languages / runtimes / frameworks / libraries (name + version constraint)
  2. Framework- or tool-mandated paths and file names only
  3. Coding conventions specific to the chosen stack
  4. Testing rules (tools, naming; required coverage only if coverage is used)
  5. Configuration rules (settings, config files, secrets handling; write "None" if not configurable)
  6. Build, packaging, and release rules (targets, build, distribution steps)
  7. Security rules tied to the chosen stack
  8. Non-goals (technologies and tools explicitly excluded)
- Flat terse bullets. "MUST / MUST NOT" phrasing. No prose, no explanations, no "why".
- Audience: AI implementer.
