Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Recall the reasoning behind a past change by locating the Claude Code transcript that produced it. Use when the user asks to "recall reasoning", "find reasoning", "look up reasoning", "recall implementation reasoning", "find the rationale", "why did I do X", "recall from transcripts", or "find the transcript for this commit".
.claude/skills/tobihagemann-recall-reasoning/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | 17% | 0% |
| case-23 | ✗→✓ | ▲ Improved | -31% | 0% |
| case-20 | ✓→✗ | ▼ Worse | -69% | 0% |
| case-21 | ✓→✗ | ▼ Worse | -1% | 0% |
| case-22 | ✓→✗ | ▼ Worse | 34% | 0% |
Recover the reasoning behind a change. Prefer Codex session history when it can be found; otherwise derive the explanation from git history and current code.
Accept any of:
<path>:<line>)If only a file is given, use git blame to resolve the commit that last touched the line.
Resolve the target commit:
bashgit rev-parse <sha> git blame -L <line>,<line> -- <path>
Collect the commit subject, changed files, and relevant diff:
bashgit show --stat --oneline <sha> git show -- <path>
If Codex session files are available, search them for the commit SHA, touched file paths, branch name, and distinctive user request text:
Write the patterns to a file with apply_patch, one per line, then match them literally:
bashrg -F -f <pattern-file> ~/.codex/sessions
-F is required, not just safer: an unescaped $ inside request text is a regex anchor and silently drops the match.
Read only the smallest relevant transcript excerpts. Prefer sessions close to the commit time and sessions that mention both the file and the task.
If no matching session is found, continue with the fallback path.
If session reasoning was found:
If no session reasoning was found:
When session reasoning was found:
markdown**Commit:** <short-sha> — <subject> **Session:** <session reference> <one or two paragraphs of reasoning>
When no session reasoning was found:
markdown**Commit:** <short-sha> — <subject> **Session:** none found <fallback explanation derived from git history and current code>
Then call update_plan to mark this step completed and continue with the next step of the active workflow.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 4,660 | 5,939 | +27% | 1 | 1 | 0% | 686 | 831 | +21% | 0 | 0 | — |
case-02 | fail→fail | 2,594 | 4,338 | +67% | 1 | 1 | 0% | 397 | 768 | +93% | 0 | 0 | — |
case-03 | fail→fail | 11,584 | 6,169 | -47% | 1 | 1 | 0% | 1,857 | 868 | -53% | 0 | 0 | — |
case-04 | pass→pass | 2,842 | 2,041 | -28% | 1 | 1 | 0% | 422 | 884 | +109% | 0 | 0 | — |
case-05 | pass→pass | 3,838 | 2,677 | -30% | 1 | 1 | 0% | 633 | 1,020 | +61% | 0 | 0 | — |
case-06 | pass→pass | 7,621 | 2,264 | -70% | 1 | 1 | 0% | 1,169 | 893 | -24% | 0 | 0 | — |
case-07 | pass→pass | 4,510 | 2,299 | -49% | 1 | 1 | 0% | 819 | 930 | +14% | 0 | 0 | — |
case-08 | fail→pass | 7,388 | 1,994 | -73% | 1 | 1 | 0% | 688 | 807 | +17% | 0 | 0 | — |
case-09 | pass→pass | 11,575 | 2,811 | -76% | 1 | 1 | 0% | 1,801 | 1,042 | -42% | 0 | 0 | — |
case-10 | pass→pass | 7,912 | 2,484 | -69% | 1 | 1 | 0% | 1,229 | 955 | -22% | 0 | 0 | — |
case-11 | fail→fail | 8,119 | 1,659 | -80% | 1 | 1 | 0% | 1,200 | 794 | -34% | 0 | 0 | — |
case-12 | pass→pass | 11,477 | 3,904 | -66% | 1 | 1 | 0% | 1,892 | 1,179 | -38% | 0 | 0 | — |
case-13 | pass→pass | 13,521 | 4,684 | -65% | 1 | 1 | 0% | 2,054 | 1,263 | -39% | 0 | 0 | — |
case-14 | pass→pass | 5,642 | 1,857 | -67% | 1 | 1 | 0% | 897 | 871 | -3% | 0 | 0 | — |
case-15 | fail→fail | 5,138 | 1,753 | -66% | 1 | 1 | 0% | 839 | 840 | +0% | 0 | 0 | — |
case-16 | pass→pass | 8,304 | 2,696 | -68% | 1 | 1 | 0% | 1,281 | 986 | -23% | 0 | 0 | — |
case-17 | pass→pass | 16,054 | 3,088 | -81% | 1 | 1 | 0% | 2,422 | 993 | -59% | 0 | 0 | — |
case-18 | pass→pass | 6,410 | 2,006 | -69% | 1 | 1 | 0% | 1,191 | 908 | -24% | 0 | 0 | — |
case-19 | pass→pass | 7,170 | 2,843 | -60% | 1 | 1 | 0% | 1,163 | 1,080 | -7% | 0 | 0 | — |
case-20 | pass→fail | 13,719 | 5,262 | -62% | 1 | 1 | 0% | 2,569 | 805 | -69% | 0 | 0 | — |
case-21 | pass→fail | 4,896 | 7,365 | +50% | 1 | 1 | 0% | 782 | 772 | -1% | 0 | 0 | — |
case-22 | pass→fail | 5,591 | 9,936 | +78% | 1 | 1 | 0% | 973 | 1,299 | +34% | 0 | 0 | — |
case-23 | fail→pass | 7,706 | 1,505 | -80% | 1 | 1 | 0% | 1,175 | 816 | -31% | 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. 23 cases were attempted, and 17 counted toward the lift figure. The other 6 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 -4 percentage points is the difference between those two pass rates over the 17 comparable cases. 4 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.