Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when composing the task brief and acceptance criteria for a Codex dispatch (invoked from dispatching-to-codex). Covers the self-contained brief format and how to write acceptance checks that actually protect you — weak checks are the sneakiest way a dispatch goes wrong.
.claude/skills/ccplugins-writing-codex-briefs/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -13% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 108% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 7% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -19% | 0% |
Codex sees zero chat history. The brief is the only channel. Anything not in it does not exist for Codex.
.codex/ACCEPTANCE.mdWritten and committed BEFORE dispatch. This is the frozen bar the judge will enforce.
GOAL: <one sentence: what "done" means>
CHECKS (all must pass, exact commands):
- `<command>` exits 0 and prints <expected>
- `<build/typecheck command>` exits 0
- `<specific test command>` exits 0
CONSTRAINTS:
- Only edit <files/dirs>. Do not modify <test files / this file / unrelated code>.Writing checks that protect you (the judge is only as strong as these):
GOAL: <same one sentence — must match ACCEPTANCE.md>
CONTEXT: <repo path, key files, constraints — everything Codex needs, it cannot ask you>
DO: <numbered, concrete steps>
VERIFY: <run the exact CHECKS above and show their real output>
REPORT: <what to return: diff summary + which checks passed>
DO NOT: commit anything; touch .codex/ACCEPTANCE.md; edit unrelated files;
delete or weaken tests; add silent fallbacks or fake/stubbed results.The DO NOT block is mandatory and always includes those five items — they are the exact ways executor models paper over failure.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 21,766 | 10,842 | -50% | 1 | 1 | 0% | 2,491 | 2,409 | -3% | 0 | 0 | — |
case-02 | fail→pass | 38,108 | 9,945 | -74% | 1 | 1 | 0% | 2,778 | 2,425 | -13% | 0 | 0 | — |
case-03 | fail→pass | 32,398 | 9,152 | -72% | 1 | 1 | 0% | 989 | 2,054 | +108% | 0 | 0 | — |
case-04 | fail→pass | 11,507 | 7,776 | -32% | 1 | 1 | 0% | 1,672 | 1,797 | +7% | 0 | 0 | — |
case-05 | fail→pass | 14,183 | 7,691 | -46% | 1 | 1 | 0% | 2,130 | 1,723 | -19% | 0 | 0 | — |
case-06 | fail→pass | 9,481 | 4,657 | -51% | 1 | 1 | 0% | 1,465 | 1,289 | -12% | 0 | 0 | — |
case-07 | fail→pass | 12,591 | 5,045 | -60% | 1 | 1 | 0% | 1,804 | 1,374 | -24% | 0 | 0 | — |
case-08 | pass→pass | 12,722 | 10,536 | -17% | 1 | 1 | 0% | 1,464 | 2,198 | +50% | 0 | 0 | — |
case-09 | fail→pass | 8,517 | 6,264 | -26% | 1 | 1 | 0% | 1,104 | 1,249 | +13% | 0 | 0 | — |
case-10 | pass→pass | 20,680 | 5,177 | -75% | 1 | 1 | 0% | 1,507 | 1,360 | -10% | 0 | 0 | — |
case-11 | fail→pass | 9,259 | 6,286 | -32% | 1 | 1 | 0% | 1,403 | 1,387 | -1% | 0 | 0 | — |
case-12 | fail→pass | 11,146 | 2,897 | -74% | 1 | 1 | 0% | 1,533 | 959 | -37% | 0 | 0 | — |
case-13 | fail→pass | 30,593 | 3,017 | -90% | 1 | 1 | 0% | 2,272 | 834 | -63% | 0 | 0 | — |
case-14 | pass→pass | 12,349 | 4,008 | -68% | 1 | 1 | 0% | 1,805 | 976 | -46% | 0 | 0 | — |
case-15 | fail→pass | 12,731 | 3,009 | -76% | 1 | 1 | 0% | 1,726 | 958 | -44% | 0 | 0 | — |
case-16 | pass→pass | 11,340 | 5,771 | -49% | 1 | 1 | 0% | 1,595 | 1,347 | -16% | 0 | 0 | — |
case-17 | pass→pass | 12,364 | 6,404 | -48% | 1 | 1 | 0% | 1,862 | 1,433 | -23% | 0 | 0 | — |
case-18 | fail→pass | 10,345 | 5,385 | -48% | 1 | 1 | 0% | 1,755 | 1,364 | -22% | 0 | 0 | — |
case-19 | fail→pass | 14,706 | 10,018 | -32% | 1 | 1 | 0% | 2,408 | 2,150 | -11% | 0 | 0 | — |
case-20 | pass→pass | 7,621 | 12,621 | +66% | 1 | 1 | 0% | 1,177 | 2,837 | +141% | 0 | 0 | — |
case-21 | fail→fail | 10,936 | 8,175 | -25% | 1 | 1 | 0% | 1,947 | 1,847 | -5% | 0 | 0 | — |
case-22 | pass→pass | 9,658 | 6,367 | -34% | 1 | 1 | 0% | 1,541 | 1,474 | -4% | 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 +64 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.