memory
Use Headroom persistent memory APIs, memory CLI, MCP/CCR retrieval, learning, verbosity, and Codex recovery.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- memory
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use Headroom persistent memory APIs, memory CLI, MCP/CCR retrieval, learning, verbosity, and Codex recovery.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/api-reference.md9.3 KB
- 参考文档references/cli-and-mcp-reference.md12.9 KB
- 参考文档references/troubleshooting.md6.2 KB
- 参考文档references/workflows.md8.6 KB
- 脚本scripts/memory_smoke.py9.9 KB
03 SKILL.md 正文
Headroom memory sub-skill
Use this sub-skill when the task is about persistent memory, memory storage maintenance, MCP tools, CCR retrieval, learning from agent transcripts, output verbosity learning, or recovery of Codex state left by Headroom wrappers.
Route here for
- Python memory APIs:
headroom.memory.Memory,with_memory,with_memory_tools,MemoryConfig,HierarchicalMemory,LocalBackend, and local/service-backed memory choices. headroom memory ...database inspection and maintenance: list, show, stats, edit, repair supersession, delete, prune, purge, export, and import.headroom mcp ...server install, uninstall, status, and serve workflows, including MCP registry behavior across supported coding agents.- CCR retrieve behavior:
headroom_retrieve,headroom_compress,headroom_stats, proxy-backed retrieval, expired hashes, and MCP tool namespace display. headroom learnfailure learning andheadroom learn --verbosityoutput-shaping calibration.headroom recover codexwhen Codex sessions/config were left in temporary Headroom homes.
Route elsewhere
- Proxy base URL setup, provider routing,
headroom proxy,headroom wrap, andheadroom unwrap: useproxy-wrapexcept for the--proxy-urlvalue passed intoheadroom mcp installorheadroom mcp serve. - Generic installation, update, deployment lifecycle,
doctor, savings dashboards, and broad operator diagnostics: useops. - Non-memory SDK compression APIs, TypeScript SDK basics, image/relevance/spreadsheet helpers, and general
HeadroomClientusage: usesdk.
Fast start
- Pick the surface:
- Local app memory API: read API reference.
- Memory database maintenance or MCP registration: read CLI and MCP reference.
- End-to-end recipes: read workflows.
- Failures: read troubleshooting.
- Keep memory state scoped. Prefer an explicit temp or app-owned
db_pathfor experiments; do not write to arbitrary project roots unless the user chose that location. - For any destructive CLI operation, preview first where supported (
--dry-run, omit--apply, or omit--yes) and confirm the target database/home. - For MCP + CCR issues, distinguish the MCP process from the proxy. The MCP server can compress locally, but proxy-backed retrieval needs a reachable proxy URL.
Bundled helper
scripts/memory_smoke.pyruns a no-credential local memory save/search/delete smoke. It uses a temporary SQLite database by default and accepts--db-pathfor a caller-chosen database and--jsonfor machine-readable output.
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).