メニュー
[Auto-run] MUST invoke before reporting code changes as complete. Trigger: any turn where you created or modified .rs files in the sage codebase. No exceptions — run even if tests pass and formatting is clean.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
[Auto-run] MUST invoke before reporting code changes as complete. Trigger: any turn where you created or modified .rs files. No exceptions — run even if tests pass and formatting is clean.
Guide to rustc's next-generation trait solver architecture (rustc_next_trait_solver + rustc_type_ir). Use when designing or implementing trait/impl resolution, type normalization, where-clause proving, or evaluating whether to integrate vs. reimplement the solver. Covers the full model: Interner trait, SolverDelegate, EvalCtxt, candidate assembly, canonicalization, search graph cycle detection, associated type normalization, and the NormalizesTo/AliasRelate split. Use whenever the user is working on trait solving, obligation discharge, or coherence checking — especially in the context of building an alternative Rust frontend or analysis tool.
[Critical] Best practice for Rust coding. Always activate this skill before authoring Rust code or answering questions about Rust.
Guide to salsa 0.26 incremental computation framework. Use when designing or implementing Salsa-based databases, tracked structs, interned types, tracked functions, or accumulators. Covers the full API: #[salsa::input], #[salsa::tracked], #[salsa::interned], #[salsa::db], #[salsa::accumulator], field tracking, the Update trait, lifecycle of tracked structs, and performance tuning. Use whenever the user is working with salsa or designing an incremental IR.
Guide to rust-analyzer's public API crates (ra_ap_*). Use this skill when building Rust analysis tools, code intelligence, linters, refactoring tools, or any programmatic consumer of rust-analyzer's libraries. Covers the full crate hierarchy: ide (Analysis/AnalysisHost), hir (Semantics, semantic model), syntax (CST/AST/parsing), ide_db (RootDatabase/Salsa), vfs (virtual filesystem), load-cargo (bootstrapping), project-model (Cargo discovery), and secondary crates (completion, SSR, hir_def, hir_ty, cfg, parser, paths, proc_macro_api). All crates at version 0.0.330. Do NOT use this skill for contributing to rust-analyzer itself or for rustc_driver-based tools (see the rustc-driver-frontend skill instead).
Use this skill when building an alternate Rust compiler frontend, analyzer, or tool that needs to consume dependency metadata (types, traits, impls, MIR) from Rust crates without decoding rlibs by hand. Covers the rustc_driver custom-driver pattern, hooking rustc_interface callbacks, extracting TyCtxt data for foreign crates, and architectural choices around Salsa interning vs arena storage for solver types. Use whenever the user is designing a Rust compiler, type checker, verifier, or analysis tool that needs rustc's view of dependencies — especially if they mention rlibs, rmeta, cargo workspaces, the next trait solver, rustc_public, stable_mir, rustc_type_ir, or consuming Rust crate metadata. Do NOT use this skill for writing plain rustc plugins like Clippy lints (standard rustc_driver custom-driver docs suffice) or for rust-analyzer contributions (rust-analyzer has its own architecture docs).
| name | sage-sanity-check |
| description | [Auto-run] MUST invoke before reporting code changes as complete. Trigger: any turn where you created or modified .rs files in the sage codebase. No exceptions — run even if tests pass and formatting is clean. |
Spawn a subagent to check for SAGE-specific anti-patterns. Pass it:
references/patterns.md found in this skill's directory.Example invocation:
Agent({
description: "Sage sanity check",
prompt: `Review the following code changes for SAGE-specific anti-patterns.
Changed files: <list the modified/new files>
Purpose: <brief description of what was done>
Follow the instructions in .agents/skills/sage-sanity-check/references/patterns.md exactly. Read that file first, then check each pattern against the changed code. Report findings in the format specified.`
})
Do NOT attempt the checks yourself — delegate to the subagent so findings come from a fresh perspective without context bias from the generation pass.