Install any skill in seconds. Free to start, no credit card required.
Get Started Free →How to think through work, distilled from a stronger model's demonstrated moves - invoke when STARTING to investigate a confusing behavior or debug a root cause; design or evaluate an approach/proposal; plan a multi-step or long-running task; verify work, review findings, or reconcile conflicting reports; write rules, reports, or docs someone else must act on.
.claude/skills/telagod-methods/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-14 | ✗→✓ | ▲ Improved | -13% | 0% |
| case-21 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-18 | ✓→✗ | ▼ Worse | -72% | 0% |
| case-04 | ✓→✓ | = Same ✓ | 11% | 0% |
| case-05 | ✓→✓ | = Same ✓ | 11% | 0% |
Five domains, each distilled from moves a stronger model actually made (the examples cite real founding-session events; most are verifiable in the repo — git log, the doctrine files, archive/ — and the rest come from the founding session's own recorded account). Governance — whether to delegate, escalate, stop, or claim done — lives in the doctrine skill; this bundle is how to work the problem once you're in it.
| You are about to… | Read (in this folder) | |---|---| | Investigate a bug, a confusing behavior, an unfamiliar system, an anomaly | investigate.md | | Propose, design, or evaluate an approach (incl. someone else's suggestion) | design.md | | Start a task with more than ~3 steps, or one that could be interrupted | execute.md | | Check work — yours or a reviewer's; weigh conflicting reports | verify.md | | Write anything another model or person must act on (rules, reports, docs) | communicate.md |
Read the ONE file that matches; each is self-contained. If two match (e.g. designing a verification), read both — they're short. When a task fires this bundle AND doctrine (e.g. "review this PR" is both a delegation decision and a verification method): doctrine FIRST — decide whether and to whom to delegate — then the matching methods file for how to do the work.
Every belief is either verified this session or hearsay — the labeling discipline and the rest of the stance live in investigate.md §3, which is their single home.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 16,236 | 14,962 | -8% | 1 | 1 | 0% | 2,480 | 2,411 | -3% | 0 | 0 | — |
case-02 | fail→fail | 14,583 | 5,422 | -63% | 1 | 1 | 0% | 2,072 | 684 | -67% | 0 | 0 | — |
case-03 | fail→fail | 29,052 | 22,570 | -22% | 1 | 1 | 0% | 4,363 | 3,995 | -8% | 0 | 0 | — |
case-04 | pass→pass | 13,882 | 12,584 | -9% | 1 | 1 | 0% | 2,065 | 2,292 | +11% | 0 | 0 | — |
case-05 | pass→pass | 12,015 | 11,029 | -8% | 1 | 1 | 0% | 1,923 | 2,136 | +11% | 0 | 0 | — |
case-18 | pass→fail | 18,433 | 15,680 | -15% | 1 | 1 | 0% | 2,061 | 575 | -72% | 0 | 0 | — |
case-06 | pass→pass | 17,191 | 16,570 | -4% | 1 | 1 | 0% | 2,570 | 2,966 | +15% | 0 | 0 | — |
case-07 | fail→fail | 20,447 | 21,290 | +4% | 1 | 1 | 0% | 2,990 | 3,211 | +7% | 0 | 0 | — |
case-08 | pass→pass | 19,130 | 16,900 | -12% | 1 | 1 | 0% | 2,771 | 2,690 | -3% | 0 | 0 | — |
case-09 | fail→fail | 19,114 | 19,364 | +1% | 1 | 1 | 0% | 3,161 | 3,297 | +4% | 0 | 0 | — |
case-10 | pass→pass | 14,170 | 16,401 | +16% | 1 | 1 | 0% | 2,137 | 2,753 | +29% | 0 | 0 | — |
case-11 | pass→pass | 19,842 | 18,829 | -5% | 1 | 1 | 0% | 3,421 | 3,644 | +7% | 0 | 0 | — |
case-12 | pass→pass | 19,076 | 17,897 | -6% | 1 | 1 | 0% | 2,851 | 2,894 | +2% | 0 | 0 | — |
case-13 | pass→pass | 16,236 | 17,104 | +5% | 1 | 1 | 0% | 2,321 | 2,836 | +22% | 0 | 0 | — |
case-14 | fail→pass | 18,182 | 13,333 | -27% | 1 | 1 | 0% | 2,661 | 2,315 | -13% | 0 | 0 | — |
case-15 | fail→fail | 17,598 | 17,413 | -1% | 1 | 1 | 0% | 2,498 | 2,866 | +15% | 0 | 0 | — |
case-16 | pass→pass | 15,379 | 13,228 | -14% | 1 | 1 | 0% | 2,459 | 2,187 | -11% | 0 | 0 | — |
case-17 | pass→pass | 18,328 | 15,309 | -16% | 1 | 1 | 0% | 2,665 | 2,592 | -3% | 0 | 0 | — |
case-19 | fail→fail | 19,676 | 16,254 | -17% | 1 | 1 | 0% | 2,799 | 2,824 | +1% | 0 | 0 | — |
case-20 | pass→pass | 18,645 | 18,720 | +0% | 1 | 1 | 0% | 2,856 | 3,213 | +13% | 0 | 0 | — |
case-21 | fail→pass | 21,397 | 21,002 | -2% | 1 | 1 | 0% | 3,199 | 3,493 | +9% | 0 | 0 | — |
case-22 | fail→fail | 11,403 | 3,626 | -68% | 1 | 1 | 0% | 1,593 | 806 | -49% | 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 20 counted toward the lift figure. The other 2 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 +5 percentage points is the difference between those two pass rates over the 20 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.