菜单
Triage and maintain TSZ diagnostic conformance. Use when investigating conformance regressions, accepted-regression drift, fingerprint-only failures, issue creation from conformance data, or focused parity fixes that must preserve the conformance gate.
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
Debug a Rust program or failing test with rdbg — set breakpoints (line, function, conditional, hit-count, panic, or watchpoint), run and step, read locals as real Rust values, change a variable mid-run, and jump to definition/hover/references via rust-analyzer. In tsz, reach for it on a runtime question reading hasn't localized — above all a wrong or extra (false-positive) diagnostic, where you break the diagnostic sinks and backtrace to the deciding frame instead of grepping emit sites. Skip it for missing-diagnostic bugs (nothing fires to break on) and small localized bugs a quick read already pins down.
Debug a Rust program or failing test with rdbg — set breakpoints (line, function, conditional, hit-count, panic, or watchpoint), run and step, read locals as real Rust values, change a variable mid-run, and jump to definition/hover/references via rust-analyzer. In tsz, reach for it on a runtime question reading hasn't localized — above all a wrong or extra (false-positive) diagnostic, where you break the diagnostic sinks and backtrace to the deciding frame instead of grepping emit sites. Skip it for missing-diagnostic bugs (nothing fires to break on) and small localized bugs a quick read already pins down.
Drive TSZ GitHub PRs through review, CI, and the merge queue. Use when checking PR status, debugging failing GitHub Actions, deciding whether a PR is WIP, marking draft PRs ready, enqueueing or verifying the queue, landing a PR, or interpreting TSZ CI gates.
Create, refresh, and publish TSZ pull requests with correct body and state. Use when opening a PR, updating a PR body, changing WIP/draft/ready state, acknowledging review, or making sure a TSZ PR satisfies the Goal, Verification, Provenance, and overlap rules.
Run a deep DRY + code-hygiene audit of the Rust workspace and turn the findings into verified, deduplicated, hierarchical GitHub tech-debt issues. Use this whenever the user asks to find duplication, code smells, derive/boilerplate bloat, oversized files, or "tech debt"; to tighten the Clippy/lint posture; to mass-file or triage hygiene issues; or to assign Priority/Effort issue Fields across the backlog. Reach for it even when the user only says "clean up the codebase", "find DRY violations", "what should we refactor", or "make issues for the cruft" — it is the right tool for evidence-backed, non-spammy hygiene issue creation.
Use when planning, implementing, reviewing, or interpreting TSZ performance work, including benchmark regressions, cache/residency changes, timing claims, perf counters, hotspot investigations, OOM/timeout/stack-overflow blockers, or optimization PR evidence.
| name | tsz-conformance |
| description | Triage and maintain TSZ diagnostic conformance. Use when investigating conformance regressions, accepted-regression drift, fingerprint-only failures, issue creation from conformance data, or focused parity fixes that must preserve the conformance gate. |
Conformance is a regression gate. Prefer checked-in artifacts and narrow filters; CI owns broad runs.
AGENTS.md and docs/plan/ROADMAP.md for conformance-affecting work.python3 scripts/conformance/query-conformance.py --dashboard
python3 scripts/conformance/query-conformance.py --campaigns
python3 scripts/conformance/query-conformance.py --fingerprint-only
python3 scripts/conformance/query-conformance.py --code TS2322
python3 scripts/conformance/query-conformance.py --code TS2322 --paths-only
Artifacts: conformance-detail.json, conformance-snapshot.json,
conformance-accepted-regressions.txt, conformance-shard-weights.json.
./scripts/conformance/conformance.sh run --filter "<name>" --verbose
Keep filters precise. Let harnesses rebuild stale binaries when possible.
Classify: new/accepted/resolved/fingerprint-only/wrong-code/missing-code/ extra-code/crash/timeout/OOM. Then identify owner: relation, inference, narrowing, indexed/keyof/mapped/conditional/template, symbol resolution, diagnostic display, parser recovery, or emit-only.
Before coding, state the structural rule and adjacent cases. Behavior changes need owning-crate tests.
Accepted-regression drift: verify shard artifacts, update the accepted file only to match observed failing set, link/file issues for new accepts, and comment with numbers plus a provenance line.