proxy-wrap
Run the Headroom proxy, route LLM providers through it, and wrap or unwrap coding agents safely.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- proxy-wrap
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Run the Headroom proxy, route LLM providers through it, and wrap or unwrap coding agents safely.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/agent-routing.md2.5 KB
- 参考文档references/proxy-reference.md2.9 KB
- 参考文档references/troubleshooting.md3.6 KB
- 脚本scripts/proxy_livez_check.py3.3 KB
03 SKILL.md 正文
Headroom proxy and agent wrapping
Use this sub-skill when the user needs a live Headroom proxy, provider routing, agent wrapping, or troubleshooting for requests that should pass through Headroom before reaching an LLM provider.
Route here for
headroom proxyforeground server flags, optimization modes, local endpoints, dashboard, health, metrics, and provider backends.headroom wrap ...andheadroom unwrap ...for Claude, Codex, Copilot, VS Code, Aider, OpenClaude, Vibe, Kimi, Grok, Cursor, Grok Build, Cline, Continue, Goose, OpenHands, OpenClaw, OpenCode, OMP, and ZCode.- Provider-specific base URL, auth, and model routing issues for Anthropic, OpenAI-compatible clients, Codex/ChatGPT backend, GitHub Copilot, Bedrock, Vertex, any-LLM, LiteLLM, and plugin integrations.
- OpenClaw/OpenCode plugin setup and proxy auto-detection.
- Proxy cache-safety, WebSocket/SSE behavior, and Prometheus metrics interpretation.
Route elsewhere
- Durable deployment lifecycle with
headroom deployorheadroom install: use../ops/SKILL.md. headroom mcp,headroom memory,headroom learn, and CCR tool registration: use../memory/SKILL.md.- Direct
compress()or TypeScript SDK application integration that does not need wrapping: use../sdk/SKILL.md.
Fast start
Foreground proxy:
headroom proxy --host 127.0.0.1 --port 8787
Claude Code through a local proxy:
ANTHROPIC_BASE_URL=http://127.0.0.1:8787 claude
# or durable wrapper:
headroom wrap claude
OpenAI-compatible client through a local proxy:
export OPENAI_BASE_URL=http://127.0.0.1:8787/v1
Check a running proxy without sending model traffic:
python scripts/proxy_livez_check.py --url http://127.0.0.1:8787 --json
References and helper
references/proxy-reference.mdcovers proxy flags, modes, endpoints, telemetry, and backend routing concepts.references/agent-routing.mdmaps agent wrappers and provider-specific routing signals.references/troubleshooting.mdcovers common proxy, base URL, auth, cloud, plugin, and cache-safety failures.scripts/proxy_livez_check.pyis a safe loopback health checker for/livez,/health, and optional/stats.
Safety rules
- Do not run long-lived proxies or durable wrappers unless the user wants that side effect; for diagnosis, use read-only health probes first.
- Do not bind the proxy to non-loopback hosts without explicit user intent and network/security context.
- Do not put API keys, OAuth tokens, cloud credentials, or bearer tokens in shell examples.
- Cloud backends such as Bedrock and Vertex require valid credentials and permissions; local help and synthetic tests do not prove live cloud access.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).