Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Keshav's second pass — a careful full read (ignoring proof/derivation detail) producing prose-level understanding sufficient to explain the paper's main contribution and evidence to a colleague. Use this after first-pass-skim, as the main content-grasping pass of the Keshav three-pass method; do not force its output into a structured data schema.
.claude/skills/yogsoth-ai-second-pass-grasp/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -37% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 41% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 47% | 0% |
Keshav's second pass: full read, proofs/derivations deferred, output is accumulated prose understanding (not a structured artifact — this is the one method in this package's whole method set that deliberately does not produce a persisted structured object).
Subagent — spawned via spawn-agent skill.
Full-text reading toward a genuine "could explain this to a colleague" understanding benefits from an uninterrupted context, distinct from the shallow first pass and the exhaustive third pass.
An earlier version of this package (staged/wechat-article-v1/skills/second-pass-grasp/) produced a draft_bundle + uncertain_fields structure for its own WeChat-article pipeline. That was correct for v1's purpose but is NOT Keshav's original second pass — this v2 SOP's output is prose, per the graph's explicit correction (context/2026-08-07-13-42-sop-pipeline-graph.html, node second-pass-grasp, "S2修订"). Do not reintroduce the bundle schema here.
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | spawn-agent | Spawn a customized CC subagent with full MCP tool access. |
<!-- END available-tables (generated) -->
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 7,668 | 32,892 | +329% | 1 | 1 | 0% | 438 | 4,626 | +956% | 0 | 0 | — |
case-02 | fail→fail | 7,729 | 34,941 | +352% | 1 | 1 | 0% | 386 | 2,388 | +519% | 0 | 0 | — |
case-03 | fail→fail | 27,067 | 26,115 | -4% | 1 | 1 | 0% | 3,838 | 3,380 | -12% | 0 | 0 | — |
case-04 | fail→pass | 42,722 | 33,942 | -21% | 1 | 1 | 0% | 8,228 | 5,179 | -37% | 0 | 0 | — |
case-05 | fail→pass | 19,639 | 23,525 | +20% | 1 | 1 | 0% | 2,542 | 3,591 | +41% | 0 | 0 | — |
case-06 | fail→fail | 8,316 | 18,747 | +125% | 1 | 1 | 0% | 375 | 858 | +129% | 0 | 0 | — |
case-07 | fail→pass | 42,828 | 44,050 | +3% | 1 | 1 | 0% | 8,213 | 7,988 | -3% | 0 | 0 | — |
case-08 | fail→fail | 37,762 | 46,079 | +22% | 1 | 1 | 0% | 6,188 | 7,745 | +25% | 0 | 0 | — |
case-09 | pass→pass | 20,164 | 21,666 | +7% | 1 | 1 | 0% | 2,389 | 2,806 | +17% | 0 | 0 | — |
case-10 | fail→fail | 32,891 | 22,867 | -30% | 1 | 1 | 0% | 4,851 | 3,252 | -33% | 0 | 0 | — |
case-11 | fail→pass | 48,467 | 43,093 | -11% | 1 | 1 | 0% | 8,228 | 6,765 | -18% | 0 | 0 | — |
case-12 | fail→pass | 21,683 | 30,215 | +39% | 1 | 1 | 0% | 2,942 | 4,336 | +47% | 0 | 0 | — |
case-13 | fail→pass | 15,552 | 19,864 | +28% | 1 | 1 | 0% | 1,546 | 2,589 | +67% | 0 | 0 | — |
case-14 | fail→fail | 23,544 | 11,702 | -50% | 1 | 1 | 0% | 1,557 | 1,113 | -29% | 0 | 0 | — |
case-15 | fail→pass | 37,778 | 23,769 | -37% | 1 | 1 | 0% | 5,499 | 3,300 | -40% | 0 | 0 | — |
case-16 | fail→fail | 28,445 | 22,588 | -21% | 1 | 1 | 0% | 3,801 | 3,183 | -16% | 0 | 0 | — |
case-17 | fail→pass | 35,080 | 33,679 | -4% | 1 | 1 | 0% | 6,237 | 4,699 | -25% | 0 | 0 | — |
case-18 | fail→pass | 25,468 | 27,931 | +10% | 1 | 1 | 0% | 3,508 | 4,230 | +21% | 0 | 0 | — |
case-19 | pass→pass | 38,200 | 25,670 | -33% | 1 | 1 | 0% | 5,644 | 3,810 | -32% | 0 | 0 | — |
case-20 | pass→pass | 13,366 | 33,519 | +151% | 1 | 1 | 0% | 1,380 | 5,411 | +292% | 0 | 0 | — |
case-21 | fail→pass | 13,037 | 44,014 | +238% | 1 | 1 | 0% | 1,267 | 7,824 | +518% | 0 | 0 | — |
case-22 | pass→fail | 13,471 | 15,472 | +15% | 1 | 1 | 0% | 1,742 | 1,930 | +11% | 0 | 0 | — |
case-23 | fail→fail | 32,908 | 54,402 | +65% | 1 | 1 | 0% | 5,008 | 8,951 | +79% | 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 22 counted toward the lift figure. The other 1 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 +39 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is 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.