메뉴
Validate that level-up flows are complete, auto-selected, editable, and evidenced with real LevelUpAgent output.
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/jleechanorg/claude-commands --skill level-up-validation명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
SOC 직업 분류 기준
SKILL.md 표시 중
| name | level-up-validation |
| description | Validate that level-up flows are complete, auto-selected, editable, and evidenced with real LevelUpAgent output. |
| scope | your-project.com level-up evidence |
Use this skill when judging whether a level-up flow works correctly for any D&D 5e class, subclass, multiclass, or campaign-defined custom class.
This is an evaluation protocol. Do not repair code from this skill unless the user explicitly asks for implementation.
testing_mcp/core/test_level_up_organic.py
evidence bundle.game_state.json, story transcript, server logs, and checksums.$PROJECT_ROOT/prompts/level_up_instruction.md and the LevelUpAgent final modal
override in $PROJECT_ROOT/agents.py.Fail closed. A level-up flow passes only when all of these are proven by real artifacts, not agent claims.
target_level > current_level.level_up_now.level_up_now opens the modal AND commits the new level atomically: the model
MUST emit state_updates.player_character_data.level = target_level on this
turn; the server calls apply_level_up which writes
together with . It does not advance
time, roll HP, award rewards, or resolve story actions.pcd.level = to_levellevel_up_session.status = in_progressThe first modal response after level_up_now must visibly include
Recommended package: and must be finalizable immediately.
The package must audit the character's class/subclass/custom class, current level, target level, campaign mechanics, and prior character sheet. It must account for every automatic gain and every player-selectable decision, including when applicable:
For prepared casters, always-prepared or feature-granted spells are additive. They do not satisfy the ordinary prepared-spell decision unless the class rules explicitly say so.
For custom classes, the model must use the custom class description and campaign rules. If a required custom-class rule is genuinely absent, the modal must stay active and ask for that missing mechanical input instead of silently skipping it.
Every required decision category must have a pre-selected recommendation before the player finishes. The visible text must distinguish:
finish_level_up_return_to_game must be present as the final accept option
once the recommended package is complete.
Every player-selectable recommendation must be editable through planning-block choices.
Acceptable patterns:
level_up_* choice for the
category.level_up_* change choice opens a follow-up modal with
concrete legal options for that category.Unacceptable patterns:
finish_level_up_return_to_game appears while unresolved decisions
remain.Free-form text must work for every selectable category while the level-up modal is active.
Validate at least one free-form edit in evidence, such as:
The response must remain in LevelUpAgent, preserve modal state, reflect the
updated pending selection, and keep finish_level_up_return_to_game available
when all required decisions are complete.
On finish_level_up_return_to_game:
player_character_data.For regressions involving stuck LevelUpAgent routing, stale modal flags, GOD
MODE promotions, or alternate level commits, evidence must include a reload case
from persisted Firestore state:
player_character_data.level, level_up_session.status,
level_up_session.current_level, level_up_session.target_level,
level_up_session.review_open, custom_campaign_state.level_up_in_progress,
and custom_campaign_state.level_up_pending.player_character_data.level >= level_up_session.target_level and
level_up_session.status == "available", the first gameplay streaming turn
must seal the session before routing.level_up_session.status is terminal (complete) or
is_session_active(...) is false, and the response must route to normal
gameplay rather than LevelUpAgent.Run the canonical real test when real LLM proof is requested:
cd testing_mcp && ../vpython core/test_level_up_organic.py --server http://127.0.0.1:8001
Do not use pytest collection for testing_mcp/ script suites.
Do not set mock-mode toggles for evidence-bearing testing_mcp/ runs.
Record the evidence bundle path, PR head SHA, server URL, model/provider, and pass/fail result.
Inspect raw artifacts, not just summary prose:
metadata.jsonrun.jsongame_state.jsonCompare the evidence bundle SHA to the live PR head. If stale, declare exactly which later diffs are behavior-neutral before accepting it.
For stuck-agent or orphan-session claims, verify the persisted reload boundary in section 7 with direct Firestore pre/post snapshots and selected-agent/routing evidence.
Use this structure:
Verdict: PASS | FAIL | PARTIAL
Evidence bundle:
PR head:
Model/provider:
Availability:
Recommended package completeness:
Auto-selection:
Click editing:
Free-form editing:
Finish commit:
Persisted reload/orphan boundary:
Blocking gaps: