Menü
Maintain models used for rpiv-pi skills and agents
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
Update nixpkgs-unstable reference in flake.nix to latest commit
Update OpenCode Go model rate limits and availability in chezmoi/dot_pi/agent/models.json
Maintain synchronized keybindings across Ghostty, WezTerm, and Kitty terminal configs. Use when editing any terminal emulator configuration file, adding new keybindings, or ensuring consistency across terminals.
| name | maintain-rpiv-pi-models |
| description | Maintain models used for rpiv-pi skills and agents |
| modified | 2026-07-21T17:14:15+0900 |
This skill maintains the per-user model override file at
chezmoi/dot_config/rpiv-pi/private_models.json
(chezmoi source path; installs to ~/.config/rpiv-pi/private_models.json).
That file maps AI/LLM models to rpiv-pi agents, stages, skills, and presets, overriding rpiv-pi's built-in defaults. It is the single place to adjust which models are used by the agent framework's various components.
Prefer task-specific model assignment rather than using one model everywhere. Use expensive models only for high-intelligence architecture/design work, and use lighter or specialized models for routine coding tasks.
| Task type | Preferred model |
|---|---|
| Default | aihubmix-oc/deep-deepseek-v4-flash |
| Visual/frontend design | aihubmix-am/xiaomi-mimo-v2.5 |
| Research / implementation / agent work | aihubmix-oc/deep-deepseek-v4-flash |
| Planning / code review / revise | aihubmix-am/cc-minimax-m3 |
| Validation / slice verification | aihubmix-oc/coding-xiaomi-mimo-v2.5-pro |
| Blueprint | aihubmix-oc/deep-deepseek-v4-flash |
| Architecture / design | aihubmix-oc/qwen3.8-max-preview |
| Claim verification | aihubmix-oc/coding-xiaomi-mimo-v2.5-pro |
| Commit generation | aihubmix-oc/deep-deepseek-v4-flash (ship preset: thinking: minimal) |
Use aihubmix-am/cc-minimax-m3 and aihubmix-oc/coding-xiaomi-mimo-v2.5-pro for tasks requiring rigid adherence to instructions. Both have low hallucination rates, so they are good choices when the model must follow constraints, preserve facts, and avoid inventing details.
aihubmix-am/cc-minimax-m3: planning, code review, and reviseaihubmix-oc/coding-xiaomi-mimo-v2.5-pro: validation and slice verificationUse openrouter/nex-agi/nex-n2-mini for things requiring rapid responses. It is the fastest model in the current configuration, so it is best suited for lightweight tasks such as discovery, exploration, claim verification, and quick turnaround where speed matters more than deep reasoning.
openrouter/nex-agi/nex-n2-miniUse for lightweight tasks:
skills.discoverskills.exploreaihubmix-oc/deep-deepseek-v4-flashUse for default, research, implementation, blueprint, commit, and agent support:
stages.commitskills.commitstages.researchstages.blueprintagents.codebase-analyzeragents.artifact-code-revieweragents.artifact-coverage-revieweragents.scope-tracerskills.create-handoffskills.resume-handoffskills.implementskills.blueprintskills.researchskills.browser-toolskills.hunk-reviewpresets.build.stages.commitpresets.build.stages.implementpresets.build.stages.researchpresets.build.stages.blueprintpresets.ship.stages.commit (thinking: minimal)presets.ship.stages.implementpresets.ship.stages.blueprintaihubmix-oc/qwen3.8-max-previewUse sparingly for expensive high-intelligence architecture/design work:
stages.architecture-reviewstages.designskills.designDo not use qwen3.8 for commits, routine research, planning, blueprint, validation, slice verification, claim verification, or code review unless there is a specific reason.
aihubmix-oc/coding-xiaomi-mimo-v2.5-proUse for validation, claim verification, and slice verification (all adversarial code-grounded tasks):
agents.claim-verifieragents.slice-verifierstages.validateskills.validatepresets.build.stages.validatepresets.ship.stages.validateaihubmix-am/cc-minimax-m3Use for planning, code review, and revision:
stages.code-reviewskills.code-reviewskills.planpresets.build.stages.code-reviewpresets.build.stages.reviseaihubmix-am/xiaomi-mimo-v2.5Use for visual/frontend design:
skills.frontend-designThis is the non-pro Xiaomi MIMO model. It is the cheapest model in the current configuration with image input and visual understanding, so use it for any UI, UX, or visual-related work.
skills.frontend-design on aihubmix-am/xiaomi-mimo-v2.5 for visual work.aihubmix-oc/deep-deepseek-v4-flash with minimal or no thinking for speed. Use thinking: minimal for the ship preset commit stage.aihubmix-oc/coding-xiaomi-mimo-v2.5-pro for validation and slice verification.aihubmix-am/cc-minimax-m3 for code review, planning, and revise.aihubmix-oc/deep-deepseek-v4-flash for research, implementation, handoffs, and agent support.