Team Polish

Before you start: read this whole install prompt, tell me in a few lines what this skill does and what installing it involves, and wait for my go-ahead before you do anything else. Prerequisites — INTEGRATED SYSTEM FIRST: this skill is one piece of the complete "Claude Code Game Studios" template (49 agents, 74 skills, studio templates and a studio CLAUDE.md, per the repository README). Each skill assumes that template is installed, with its conventions, its project.yaml / project.local.yaml configuration, its .claude/docs/ mode documents, its hooks (for example hooks/yaml-helper.sh, outside this skill's folder) and its .claude/scripts/ helpers. Installing this listing alone copies only this skill's folder, without the rest of the system: the skill may reference template agents, files, hooks or scripts that are absent outside the full template, and it may not run as designed on its own. No account, no key and no payment are needed to copy the files, and nothing is run during installation. Install "Team Polish" for me. The skill itself orchestrates the polish team to optimize and harden one feature or area; with no argument it prints usage and exits without spawning any agent. It first resolves review_mode, automation, team.size and workflow by running the template's hooks/yaml-helper.sh, outside this skill's folder, and announces the active set before Phase 1: at the default team.size individual the performance-analyst and technical-artist run and no programmer is active, so programmer-owned optimisation items are not implemented in that run — the report hands the list to /dev-story; at small the sound-designer, qa-tester and engine-programmer join (plus the tools-programmer when a cause traces to a content authoring tool), and at studio an adversarial QA pass is added. Phase 1 delegates profiling to the performance-analyst, who profiles the area using /perf-profile and produces a prioritised optimisation list (change, expected gain, owner per item) — the performance-analyst writes no game code. Phase 2 implements that list through its owners in the active set (engine-programmer for engine-level causes, tools-programmer for authoring-tool causes), while Phase 3 (technical-artist: VFX, particles, shaders, visual juice) and Phase 4 (sound-designer: audio completeness, mix levels, ambient layers, spatial checks) run in parallel with it; every implementing agent in Phases 2 to 4 first returns the files it will create or change, and the orchestrator asks once for the whole Phase 2-4 set before any edit, because those phases edit existing code, shaders, audio and assets and are outside the bounded write exception. Phase 5 delegates hardening to the qa-tester: edge cases, soak tests, stress tests, regression tests naming each Phase 2-4 change, and minimum-spec testing where available. Phase 6 signs off READY FOR RELEASE, NOT ASSESSED or NEEDS MORE WORK under strict rules the file spells out: the performance budgets in project.yaml (performance.target_framerate, frame_budget_ms, draw_call_limit, memory_ceiling_mb) have no defaults, so an unset budget means that metric was never compared and the verdict cannot be READY; a phase that did not run likewise yields NOT ASSESSED, and a known over-budget metric or unresolved regression yields NEEDS MORE WORK. Each agent writes one dated report file under production/polish/, a location the file states nothing in the repo reads yet — the reports are records, not gate inputs. This orchestrator writes no files directly. Its frontmatter pins model: sonnet. Repository: https://github.com/Donchitos/Claude-Code-Game-Studios (the skill is the whole .claude/skills/team-polish/ folder at the pinned commit: 1 file in total) Pinned version: install the code at commit 05ad7c8a5b8aa2aa86bbd3439ae7cf7295cac418 exactly. Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Clone https://github.com/Donchitos/Claude-Code-Game-Studios into a temporary folder, check out commit 05ad7c8a5b8aa2aa86bbd3439ae7cf7295cac418, and copy the .claude/skills/team-polish/ folder from that checkout. Then compare every file in the folder with the fingerprints recorded for this listing on 2026-10-22: [FINGERPRINTS] - SKILL.md: SHA-256 8d394fd7111b8ee96d4a91809e6959225d73938ffeafdaa06b0fa5a33679086d If a file does not match, it changed since the listing was prepared: do not install it yet, tell me what you found and let me decide. 2. Read SKILL.md in full, and tell me in plain words: what the skill does when used, what it connects to, whether it reads or sends any personal information or credentials, and anything that seems unusual. Stop on any red flag and tell me before going further. Confirm these points from the files: - The skill folder holds 1 file: SKILL.md only. Nothing in the folder is a program; it is instructions the agent follows at use time. - Nothing is sent anywhere during installation itself; the only download at install time is the skill folder from GitHub at the pinned commit, and no project of mine is audited, read or modified. - Integrated-system reality, from the file: at use time the skill runs the template's hooks/yaml-helper.sh to resolve review_mode, automation, team.size and workflow, follows .claude/docs/automation-modes.md, config-resolution.md and error-recovery-protocol.md, uses the template's /perf-profile skill, reads my project's code, art bible and performance budgets in project.yaml, and spawns template agents (performance-analyst and technical-artist at every size; engine-programmer, tools-programmer, sound-designer and qa-tester at wider sizes) — agents, docs, skills and project files that live outside this skill's folder and are absent in a standalone install, so it cannot run as designed on its own. - At use time its sub-agents write dated report artifacts under production/polish/ without a per-write prompt, but only inside the bounded exception the file documents (orchestrator-named path, new artifact under production/, docs/ or tests/, phase gated by an AskUserQuestion) — and the file states plainly that nothing in the repo reads production/polish/ yet. The optimisation and polish changes of Phases 2 to 4, which edit existing code, shaders, audio or assets, are outside that exception: each implementing agent returns its file set first and edits only after one ask for the whole Phase 2-4 set. This orchestrator writes no files directly. No third-party service, account or credential is named in the file. - Do not polish or profile any feature, do not spawn any team agent, do not run any template hook, and do not use the skill in any way during this install: using it is a separate step that only starts when I ask for it later. 3. When I say "yes, install", install it as a skill: copy the whole team-polish folder into the agent's skills directory, in a folder named exactly "team-polish", keeping its 1 file in their original relative layout (SKILL.md at its root). 4. Verify with no network calls: frontmatter valid (name: team-polish), the 1 file in place with matching fingerprints. Do not run anything from the folder and do not use the skill to verify. 5. Report what was installed, where, and what I still need to do myself (for example: that it belongs to the full Claude Code Game Studios template and may not run standalone (it resolves its settings through the template's yaml-helper hook, reads the template's project code and budgets and spawns its polish team agents), and that at use time its report artifacts are written under production/polish/ under a documented bounded exception, Phases 2 to 4 edits only after one ask for the whole set, and at the default team size programmer-owned optimisation items are handed to /dev-story instead of being implemented). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. If anything looks off, stop and ask me.