一键导入
evor-doctor
Check environment health and .evor integrity; auto-repair obvious issues
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。
菜单
Check environment health and .evor integrity; auto-repair obvious issues
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。
基于 SOC 职业分类
Mission interview workflow that elicits all mission settings and launches a validated run via evor_init_run
Start the live FastAPI + SSE dashboard for the active Evor mission on port 8756
Scan an existing ML workspace to produce a starting-point report; pre-fills the evor-setup interview
Inspect accumulated Gotchas (failures + hardware limits) and the hardware capability profile
Evor MCP tool catalog, run lifecycle, artifact schemas, and native-tool guide. Auto-loads when operating an evor run — setting up/running/resuming/inspecting a mission, recording nodes/evals, launching or watching training, reading/writing tick artifacts, citing papers, emitting/querying signals, or changing any evor state.
Generate a final Evor mission report with tree visualization, frontier table, and static HTML export
| name | evor-doctor |
| description | Check environment health and .evor integrity; auto-repair obvious issues |
| argument-hint | [run-dir] [--repair] |
| level | 2 |
| skills | ["oh-my-evor:evor-mcp"] |
<Use_When>
/evor-doctor/evor-validate reports infrastructure issues (corrupt files, missing directories)<Do_Not_Use_When>
/evor-validate/evor-run
</Do_Not_Use_When>Parse the argument for an optional run-dir path and --repair flag:
<run_dir>: run env + .evor checks for that specific run directory.--repair: pass the repair flag to auto-fix repairable issues.Call evor_doctor with the applicable arguments:
evor_doctor({}) // env-only check
evor_doctor({ run_id: "<run_dir>" }) // env + .evor integrity check
evor_doctor({ run_id: "<run_dir>", repair: true }) // env + .evor + auto-repair
Parse the JSON output and present a categorised summary:
=== evor-doctor report ===
Environment
✓ python_version — Python 3.11.4
✓ torch — torch 2.1.0 importable
✓ node — node v20.5.1
✓ env_evor_root — EVOR_ROOT not set (will use default .evor/)
✓ patch — patch found at /usr/bin/patch
.evor integrity (run: <run_dir>)
✓ run_dir — run directory found and accessible
✓ tree_json — DICT format (12 nodes)
⚠ mission_state — mission state not found — run /evor-setup
✓ orphan_pending — no orphaned pending nodes
✓ frozen_hash — frozen-splits integrity confirmed
Verdict: OK — no errors detected
Status icons: ✓ ok — check passed ⚠ warn — warning (non-blocking; action recommended) ✗ error — error (must be resolved) 🔧 repaired — issue was auto-fixed with --repair
For each ERROR item, provide the remediation:
| Error | Remediation |
|---|---|
python_version < 3.10 | Upgrade Python or activate a 3.10+ environment |
torch not importable | pip install torch or use a GPU-enabled environment |
node missing | Install Node.js (required for hooks) |
mission_state missing | Re-run /evor-setup to initialize mission state |
orphan_pending_nodes | Use /evor-doctor --repair to clear orphaned state, or resolve the pending nodes through the normal tick workflow |
frozen_split_integrity | Frozen splits were modified after locking — re-run /evor-setup |
run_dir missing | Verify the run directory path |
For WARN items:
torch warn: GPU training paths gated, but CPU-only runs still workIf --repair was passed and repairs were made, display:
Repaired:
🔧 <item>: <description of repair performed>
Re-run /evor-validate to confirm the contract is now valid.
<Tool_Usage>
evor_doctor — run environment and .evor integrity diagnostics; pass repair: true to auto-fix
</Tool_Usage>