$npx -y skills add 100yenadmin/fable-token-saving-skills-orchestrator --skill run-economicsUse when choosing between cheap probes, worker runs, expensive ruler runs, or long keepalive waits.
| 1 | # Run Economics |
| 2 | |
| 3 | Match the instrument to the question. |
| 4 | |
| 5 | ## Mechanism |
| 6 | |
| 7 | Use cheap probes when the question is "does this mechanism fire?" |
| 8 | |
| 9 | Examples: |
| 10 | |
| 11 | - hook blocks stale wait |
| 12 | - cue reaches worker |
| 13 | - small parser case passes |
| 14 | - one narrow acceptance test catches the bug |
| 15 | |
| 16 | ## Ruler |
| 17 | |
| 18 | Use premium runs when the question is "does this meet the bar?" |
| 19 | |
| 20 | Examples: |
| 21 | |
| 22 | - release readiness |
| 23 | - final quality score |
| 24 | - customer-facing proof |
| 25 | - hard architectural judgment |
| 26 | |
| 27 | ## Cache |
| 28 | |
| 29 | For short waits, preserve the prompt cache with a sleep-tick if no inline work |
| 30 | remains. For long waits, let the cache die once. Do not ping through multi-hour |
| 31 | idle windows by default. |
| 32 | |
| 33 | ## Ledger |
| 34 | |
| 35 | For scored or judged runs, record provider, model or lane, scenario, method, |
| 36 | score, and proof boundary. |