state-recovery-and-experiment-ops
Use ARIS Research Wiki, project state files, session recovery, watchdog monitoring, experiment queues, and remote/GPU operating guidance safely.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- state-recovery-and-experiment-ops
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Use ARIS Research Wiki, project state files, session recovery, watchdog monitoring, experiment queues, and remote/GPU operating guidance safely.
- disable-model-invocation
- true
02 资源文件 3
- 参考文档references/project-state.md1.6 KB
- 参考文档references/troubleshooting.md1.6 KB
- 参考文档references/watchdog-and-wiki.md1.8 KB
03 SKILL.md 正文
State Recovery and Experiment Operations
Use this sub-skill when an ARIS workflow must persist research knowledge, resume after compaction or a crashed session, initialize or query research-wiki/, monitor long-running tasks, register screen/tmux jobs, or plan local/remote/GPU experiment operations.
Route Here
- Initialize and maintain Research Wiki papers, ideas, experiments, claims, graph edges, log, and query pack.
- Resolve ARIS helpers from
.aris/tools, local tools, manifest, or pointer files. - Build recovery checkpoints from pipeline status and canonical artifacts.
- Register and monitor watchdog tasks for training and downloads.
- Hand off experiments to local/remote/Vast/Modal/GPU environments without claiming unverified hardware.
- Diagnose state drift, missing helper scripts, stale task status, and interrupted review loops.
Reroute
- Installer/manifests and host skill roots:
../install-and-distribution/SKILL.md. - Workflow/artifact selection:
../workflow-routing-and-skill-catalog/SKILL.md. - Reviewer traces, MCP state, provider credentials:
../review-and-provider-backends/SKILL.md. - Changes to helper scripts or watchdog/wiki tests:
../repository-maintenance/SKILL.md.
Recovery Pattern
- Find the project root (prefer the Git root when available).
- Read pipeline status, research contract, fixed-name handoffs,
REVIEW_STATE.json, and recent trace directories. - Check active training/download sessions and watchdog status before restarting anything.
- Reconstruct the next stage from files; never rely on memory after compaction.
- Write an explicit checkpoint before another long tool call.
Reference Map
references/project-state.mdexplains the project file contracts, priority order, and recovery sequence.references/watchdog-and-wiki.mdexplains Research Wiki commands, task JSON, status semantics, and helper lookup.references/troubleshooting.mdcovers stale state, missing helpers, dead sessions, and remote/GPU uncertainty.- Root
../../references/helper-resolution-and-project-files.mdgives the shared resolver and artifact map.
Safety
- Do not start or kill remote jobs without user intent.
- Do not treat a missing GPU probe as a successful GPU run.
- Preserve append-only logs and versioned artifacts.
- Keep API credentials and private server paths out of persistent research artifacts.
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).