菜单
Run the full Sailfin compiler validation pipeline (make clean-build then make check). Use before declaring a feature shipped, before cutting a release, or when you need to verify self-hosting still works after a structural change.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Systematically diagnose why a Sailfin source file or self-hosting build step fails to compile. Isolates the failure, identifies the pipeline stage, escalates to seed-stabilizer or compiler-architect when the fix is non-trivial, and verifies via test-runner. Use whenever `make compile` fails, a .sfn file won't build, or LLVM rejects generated IR.
Systematically diagnose Sailfin compilation, self-hosting, or LLVM lowering failures without adding build-driver workarounds.
Run Sailfin compiler and test verification safely with the required self-hosting and formatting gates.
Pick up a ready Sailfin Linear or GitHub issue and drive it through branch, implementation, verification, and PR handoff.
Plan Sailfin work in Linear from Initiatives through Projects to session-sized compiler/runtime issues. Use when Codex needs to audit or create Linear planning structure, prioritize production compiler/runtime health and performance, assign Projects/Cycles/priorities/estimates, groom Triage/Backlog/Ready/Blocked queues, or propose/apply bulk Linear issue updates.
Pin the Sailfin seed compiler to a specific released version. Updates bootstrap.toml [seed].version and the compiler [toolchain] floor, fetches the binary, and optionally runs a smoke compile to verify the new seed builds the compiler. Usage: /pin-seed <version> (e.g. /pin-seed v0.5.10-alpha.12 or /pin-seed 0.5.10-alpha.12).
| name | check |
| description | Run the full Sailfin compiler validation pipeline (make clean-build then make check). Use before declaring a feature shipped, before cutting a release, or when you need to verify self-hosting still works after a structural change. |
| allowed-tools | Bash, Read |
This skill runs make check — see CLAUDE.md ## The validation ladder for what that pipeline covers and when to reach for it instead of a cheaper rung.
.claude/skills/check/scripts/run-check.sh
The script writes a timestamped log to build/logs/check-<ts>.log, and exits non-zero on failure with the last 80 lines of the log on stdout.
If the script exits non-zero, read the log, identify the failing pipeline stage, and either:
compiler/src/*.sfn), orseed-stabilizer agent if the root cause isn't obvious.The four failure surfaces, in order of likelihood:
| Stage | Symptom | First place to look |
|---|---|---|
| Build from seed | Seed can't compile current source | Recent changes to parser/typecheck/emit |
| First-pass test suite | Behavior regression | The test's pipeline stage |
| Seedcheck build | First-pass binary miscompiles itself | LLVM lowering, emit_native |
| Hello-world smoke | Seedcheck can't run a basic program | Entry point, linker symbols |
See CLAUDE.md ## The validation ladder — this skill is the top rung; use the
cheapest rung that catches the error instead.
Expect 15–20 minutes on typical hardware. If the script runs past 30 minutes, something is wrong — investigate instead of waiting.