Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when recovering a Codex session, especially if the user provides a Codex session id or wants recent Codex sessions listed before resuming work.
.claude/skills/cnfjlhj-session-recovery-codex/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✓→✓ | ＝ Same ✓ | -17% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 24% | 0% |
| case-06 | ✓→✓ | ＝ Same ✓ | 1% | 0% |
| case-11 | ✗→✗ | ＝ Same ✗ | 51% | 0% |
| case-01 | ✗→✗ | ＝ Same ✗ | 204% | 0% |
当用户消息里已经包含形如 019b6d01-790f-7a32-9d75-932700db98a4 的 session id 时：
1) 直接运行统一入口脚本：
python3 ~/.codex/skills/session-recovery-codex/scripts/recover-session.py --id "<session_id>" --cwd "$PWD"
2) 向用户展示脚本输出，并据此复述：
3) 询问用户：
我的理解是否正确？你想继续哪部分？
1) 运行统一入口脚本扫描当前项目最近会话：
python3 ~/.codex/skills/session-recovery-codex/scripts/recover-session.py --cwd "$PWD" --limit 5
2) 将脚本输出的最近 5 个会话概览展示给用户，并让用户输入要恢复的序号（1–5）。
3) 用户选择后，拿到对应 session id，再运行统一入口脚本做精确恢复：
python3 ~/.codex/skills/session-recovery-codex/scripts/recover-session.py --id "<session_id>" --cwd "$PWD"
4) 向用户展示：
update_plan 记录恢复）5) 询问用户：“我的理解是否正确？你想继续哪部分？”并在用户确认后继续执行任务。
~/.codex/sessions 恢复事实记录。| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-11 | fail→fail | 3,310 | 4,826 | +46% | 1 | 1 | 0% | 452 | 682 | +51% | 0 | 0 | — |
case-01 | fail→fail | 5,689 | 6,452 | +13% | 1 | 1 | 0% | 256 | 779 | +204% | 0 | 0 | — |
case-02 | fail→fail | 5,522 | 4,980 | -10% | 1 | 1 | 0% | 734 | 709 | -3% | 0 | 0 | — |
case-03 | fail→fail | 14,251 | 4,812 | -66% | 1 | 1 | 0% | 1,593 | 797 | -50% | 0 | 0 | — |
case-04 | pass→pass | 11,527 | 10,728 | -7% | 1 | 1 | 0% | 2,109 | 1,758 | -17% | 0 | 0 | — |
case-05 | pass→pass | 8,578 | 9,043 | +5% | 1 | 1 | 0% | 1,603 | 1,988 | +24% | 0 | 0 | — |
case-06 | pass→pass | 9,602 | 6,919 | -28% | 1 | 1 | 0% | 1,734 | 1,745 | +1% | 0 | 0 | — |
case-07 | fail→fail | 3,089 | 6,532 | +111% | 1 | 1 | 0% | 444 | 763 | +72% | 0 | 0 | — |
case-08 | fail→fail | 4,063 | 5,015 | +23% | 1 | 1 | 0% | 502 | 682 | +36% | 0 | 0 | — |
case-09 | fail→fail | 10,985 | 6,585 | -40% | 1 | 1 | 0% | 1,769 | 905 | -49% | 0 | 0 | — |
case-10 | fail→fail | 4,494 | 6,781 | +51% | 1 | 1 | 0% | 716 | 887 | +24% | 0 | 0 | — |
case-12 | fail→fail | 10,772 | 3,938 | -63% | 1 | 1 | 0% | 1,776 | 1,013 | -43% | 0 | 0 | — |
case-13 | fail→fail | 6,281 | 7,546 | +20% | 1 | 1 | 0% | 351 | 969 | +176% | 0 | 0 | — |
case-14 | fail→fail | 9,626 | 2,021 | -79% | 1 | 1 | 0% | 1,455 | 726 | -50% | 0 | 0 | — |
case-15 | fail→fail | 7,878 | 2,463 | -69% | 1 | 1 | 0% | 1,100 | 737 | -33% | 0 | 0 | — |
case-16 | fail→fail | 10,827 | 2,878 | -73% | 1 | 1 | 0% | 1,707 | 903 | -47% | 0 | 0 | — |
case-17 | fail→fail | 6,697 | 7,445 | +11% | 1 | 1 | 0% | 227 | 871 | +284% | 0 | 0 | — |
case-18 | fail→fail | 3,341 | 5,673 | +70% | 1 | 1 | 0% | 592 | 768 | +30% | 0 | 0 | — |
case-19 | fail→fail | 10,852 | 4,009 | -63% | 1 | 1 | 0% | 1,547 | 1,076 | -30% | 0 | 0 | — |
case-20 | fail→fail | 3,236 | 6,858 | +112% | 1 | 1 | 0% | 519 | 853 | +64% | 0 | 0 | — |
case-21 | fail→fail | 4,700 | 6,118 | +30% | 1 | 1 | 0% | 690 | 1,296 | +88% | 0 | 0 | — |
case-22 | fail→fail | 3,500 | 6,135 | +75% | 1 | 1 | 0% | 457 | 802 | +75% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 9 counted toward the lift figure. The other 13 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of 0 percentage points is the difference between those two pass rates over the 9 comparable cases.
Other measured skills in the registry, with their headline benchmark lift.