Menü
Launch worker subagents that read gcsim character source code and extract v2-schema particle data. Use when the user asks to populate, update, or audit particle data from gcsim for the ER calculator.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Basierend auf der SOC-Berufsklassifikation
Reference and verification tools for the Genshin Impact damage calculator. TRIGGER when: verifying or looking up character/weapon/artifact game data (talent values, buff percentages, multipliers, conditions), reviewing or writing buff/formula implementations, or needing damage calc code patterns. Key tool: `impl_audit.py show C|W|A <id>` exports official game text and talent parameters — use this as the first step to verify in-game values.
Pull and triage GenshinTools user feedback from the production Cloudflare D1 feedback table. TRIGGER when the user asks to fetch, pull, review, summarize, or triage newest feedback, feedback since last time, feedback deltas, feedback-reports, or "what users said / what to build next". Uses the repo's incremental `npm run feedback:fetch` flow and summarizes only newly fetched submissions.
Draft and publish player-facing "What's New" / update log entries for GenshinTools. TRIGGER when the user asks to write, update, draft, or publish the update log / changelog / "what's new" / 更新日志. Gathers commits since the last published entry, summarizes only user-facing changes (no infra/refactors, no engineering tone), proposes bilingual zh+en entries for approval, then writes them to src/presets/updatelog/{en,zh}.md under the current date. Do NOT auto-run on commit or on a schedule — this is an on-demand, approval-gated flow.
UI design decision guide for GenshinTools. Use only for open-ended frontend design work: new surfaces, substantial redesigns, or unclear UX/layout choices where the agent must choose the structure. Do not use for direct user instructions, bug fixes, copy tweaks, mechanical UI changes, or routine edits where nearby code already shows the pattern.
Launch damage calculator worker subagents (review, triage, implement, excel, cleanup). Use when the user asks to review, audit, triage, implement, cross-validate, or clean up damage formulas or buff implementations.
Launch energy recovery worker subagents to populate selfEnergy data files. Use when the user asks to generate, update, or review self-energy recovery data for the ER calculator.
| name | gcsim-particle-extract |
| description | Launch worker subagents that read gcsim character source code and extract v2-schema particle data. Use when the user asks to populate, update, or audit particle data from gcsim for the ER calculator. |
Launch parallel worker subagents to extract per-character particle generation data from a local gcsim clone, and emit v2-schema batch JSON for later merge into src/data/ercalc/particles.gcsim.json.
Fandom (particles.json) and Lunaris (particles.lunaris.json) remain as-is for cross-validation. This skill produces a third, higher-fidelity reference sourced from gcsim's actual runtime logic (done flags, probability rolls, ICD gating, etc.).
Prerequisites: gcsim cloned at
F:/Codes/genshin/gcsim. Particle-location index built viauv run --project scripts/pyproject.toml scripts/gcsim_particle_locations.py→scripts/out/particle-locations.json.
The location scanner covers 99 characters (Barbara and the Traveler base file have no particle logic — handled separately). Characters released after the last gcsim sync (Flins, Varesa, Ineffa, Jahoda, Iansan, Ifa, Illuga, Nicole, Lohen, Prune, Durin, etc.) are not in gcsim and fall back to Fandom/Lunaris.
.agents/agents/gcsim-particle-extract.md contains the extraction rules. Each worker reads that file, processes its assigned batch of characters, and writes to its own batch output file.
Batch size: 15 characters per agent. For a full sweep of 99 covered chars → 7 agents in parallel.
Launch each agent with:
Read `.agents/agents/gcsim-particle-extract.md` and follow its instructions.
Batch: <N>
Characters: <id1>, <id2>, ..., <id15>
Example for a subset re-run:
Read `.agents/agents/gcsim-particle-extract.md` and follow its instructions.
Batch: 1
Characters: bennett, fischl, hutao, raiden, klee, diona, charlotte, emilie, nahida, albedo, alhaitham, xiangling, xingqiu, ayaka, ayato
All agents run concurrently. Use Codex's worker subagent type for each batch. Tell each worker that it is not alone in the codebase, must not revert edits made by others, and owns exactly its assigned batch output file.
Standard sweep of 99 chars ÷ 15/batch = 7 agents. Send all 7 in one batched message. Wall-clock ~3-5 min.
Each agent writes to scripts/out/particles.gcsim.batch_<N>.json. After all agents complete:
uv run --project scripts/pyproject.toml scripts/merge_gcsim_batches.py
This merges batches into src/data/ercalc/particles.gcsim.json, validates each entry against the v2 schema, and reports:
_unmodeled notes (for human review)particles.json (Fandom) and particles.lunaris.json (Lunaris). Expect matches for simple chars; gcsim should be more accurate for Klee-style (done-gated), Cyno-style (variant skills), and Diona-style (multi-hit) cases._unmodeled aggregation — decide which patterns warrant schema extensions, which stay curated.particles.gcsim.json to particles.json (as source: "gcsim") only after review — don't overwrite automatically.