菜单
Orientation for a session starting work on this codebase - reading order, where things live, which skill to use for which task, and the non-negotiable constraints. Use at the start of a session, when unsure where to look, or when the user asks how to approach a task here.
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。
Hostile security and robustness audit of src/ assuming malicious traffic and 100x load. Use before releases, after large feature merges, or when the user asks for a security audit, adversarial review, or "what breaks under attack".
Holistic system audit for architecture drift, Hard Rules compliance, and async integrity across recently changed code. Use after completing a batch of tasks, before a release, or when the user asks for a health check or audit.
Semantic-correctness audit of hot-path code - enumerate the invariants a component must maintain, then hunt code paths that violate them (wrong answers, not crashes). Use after changing routing, eviction, accounting, or counter logic, before a release, or when metrics drift without any crash.
Plan, run, and interpret Locust load benchmarks (mock-backend and real-GPU) including A/B routing experiments. Use when the user wants to measure cache-hit ratio, TTFT, or throughput, compare routing modes, chase a benchmark regression, or design a benchmark experiment.
Triage build or runtime failures. Use when the user pastes a compile error, link error, runtime crash, test failure, or error log, or mentions a build failure.
Post-implementation maintenance - write unit tests, sync documentation, and update BACKLOG.md. Use after code is implemented and reviewed, or when the user asks for tests or doc updates for a change.
基于 SOC 职业分类
| name | orient |
| description | Orientation for a session starting work on this codebase - reading order, where things live, which skill to use for which task, and the non-negotiable constraints. Use at the start of a session, when unsure where to look, or when the user asks how to approach a task here. |
| argument-hint | ["task-at-hand","optional"] |
Task at hand: $ARGUMENTS
Ranvier Core is a Layer 7+ LLM traffic controller: C++20 on Seastar (shared-nothing, thread-per-core). It routes inference requests by token prefix to keep GPU KV-caches warm.
cmake/make/builds in this sandbox — Seastar deps are too heavy. Verification happens in the developer's Docker container; your job is to emit exact commands (Deferred Gates, see /validate) and never claim they passed..dev-context/claude-context.md are mandatory for all new and modified code. If you have not read them this session, read them before writing C++.std::shared_ptr, async-only I/O, bounded containers. When unsure whether a Seastar API works the way you think, verify against Seastar docs — do not guess.docs/ or assets/ (token-heavy). Read specific docs/internals/*.md files by name; grep .dev-context/cheatsheet.md (2500 lines of operational runbook) rather than reading it whole..dev-context/ (dated audits, investigations) are frozen — quote them, never edit them.CLAUDE.md (auto-loaded) → .dev-context/claude-context.md — architecture, layout, conventions, Hard Rules. Always.docs/internals/<area>.md deep-dive — authoritative for its area; must be updated in the same PR that changes its area..dev-context/README.md indexes everything else (pitfalls reference, sync map, investigations, audit records).
| You are... | Use |
|---|---|
| Planning a multi-file feature | /plan |
| Implementing a feature or change | /implement |
| Adding a whole new service/subsystem | /new-service |
| Triaging a compile error, crash, or failing test | /debug-build |
| Responding to a production problem | /incident |
| Investigating performance | /perf, then /benchmark to measure |
| Designing/running a load benchmark or A/B | /benchmark |
| Choosing what tests/verification to run | /validate |
| Reviewing code before commit | /review |
| Writing tests + syncing docs after a change | /doc |
| Restructuring without behavior change | /refactor |
| Periodic health check of recent work | /audit |
| Hostile security/robustness sweep | /adversarial-audit |
| Hunting wrong-answer bugs (counter drift, side-index divergence, contract violations) | /invariant-audit |
| Turning a bug class into a Hard Rule | /extract-pattern |
| Assessing project direction | /strategic-review |
| Area | Where |
|---|---|
| Routing core (ART, scoring, learn/evict) | src/router_service.*, src/radix_tree.hpp, src/route_scorer.hpp |
| HTTP ingress/proxy/streaming | src/http_controller.*, src/stream_parser.*, src/connection_pool.hpp |
| Tokenization (3-layer) | src/tokenizer_service.*, src/tokenizer_thread_pool.* |
| Cluster gossip + DTLS | src/gossip_*, src/dtls_context.*, src/crypto_offloader.* |
| KV-event ingestion (vLLM ZMQ) | src/kv_event_{decoder,ledger,subscriber}.* |
| GIE Endpoint-Picker (Envoy ext_proc) | src/gie_epp_* |
| Telemetry / usage accounting | src/telemetry_*, src/usage_ledger_*, src/gpu_seconds_accounting.hpp |
| Persistence (SQLite, MPSC) | src/async_persistence.*, src/sqlite_persistence.*, src/mpsc_ring_buffer.hpp |
| Lifecycle & wiring | src/application.cpp (startup() ~997, stop_services() ~1736) |
| Config | src/config_*.hpp, ranvier.yaml.example |
| Unit tests / fuzz / integration | tests/unit/, tests/fuzz/, tests/integration/ |
| Work tracking | BACKLOG.md (current), CHANGELOG.md |
State the task type you've identified and invoke the matching skill. If the task spans several (most features do: plan → implement → review → doc → validate), say the sequence up front.