ops
Operate Headroom installs, durable deployments, diagnostics, savings reports, output shaping, evals, and bundled tools.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- ops
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Operate Headroom installs, durable deployments, diagnostics, savings reports, output shaping, evals, and bundled tools.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/cli-reference.md4.8 KB
- 参考文档references/troubleshooting.md4.4 KB
- 参考文档references/workflows.md3.5 KB
- 脚本scripts/diagnose_headroom_install.py5.2 KB
03 SKILL.md 正文
Headroom ops
Use this sub-skill when the user needs to install, update, deploy, diagnose, report on, or maintain a Headroom runtime without changing application code.
Read this for
headroom deployandheadroom install ...persistent proxy profiles.headroom init ...durable agent hooks/provider routing for supported agents.headroom updateand install-method guidance.headroom doctor,dashboard,inspect,perf,savings,output-savings,agent-savings,audit-reads, andcapture network-diff.headroom tools ...bundledsg,diff, andlochelpers.headroom evals ...commands and when not to run expensive or API-key-dependent evals.- Cross-cutting filesystem roots and path overrides.
Route elsewhere
headroom proxy, provider base URLs,headroom wrap, andheadroom unwrap: read../proxy-wrap/SKILL.md.headroom memory,headroom mcp,headroom learn,headroom recover codex, and CCR retrieval: read../memory/SKILL.md.- Direct Python or TypeScript SDK usage: read
../sdk/SKILL.md.
First checks
-
Verify the package and CLI:
python -c "import headroom; print(headroom.__version__)" headroom --version -
For a broad diagnosis, run:
headroom doctor python scripts/diagnose_headroom_install.py --check-cli -
For durable deployment status, use:
headroom install status --profile default -
For token/cost reporting, prefer the least invasive read-only command first:
headroom savings headroom perf --hours 24 headroom output-savings
References and helper
references/cli-reference.mdcatalogs operator-facing commands, safe checks, and side-effect boundaries.references/workflows.mdgives task-oriented install, deployment, diagnostics, reporting, tools, and eval recipes.references/troubleshooting.mdmaps common symptoms to recovery steps.scripts/diagnose_headroom_install.pyis a safe local checker for imports, CLI availability, and canonical path roots.
Safety rules
- Do not run
headroom update,headroom deploy,headroom install apply/remove,headroom init,headroom savings --reset,headroom memory purge, or any command that mutates user config/state unless the user explicitly wants that side effect. - Treat user configuration files such as Claude, Codex, OpenClaw, OpenCode, and shell startup files as user-owned; if a file is malformed, instruct the user to fix or move it rather than overwriting it.
- Do not paste API keys, bearer tokens, cloud profiles, proxy credentials, or private paths into examples.
- If a command depends on past proxy logs, savings ledgers, or a running proxy, first check whether the relevant file or endpoint exists and explain empty results as normal on a new install.
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).