Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Install, update, version-check, or uninstall the Coco open-source AI workflow framework via its CLI (cocosuperintelligence, run with npx). Use when setting up Coco on a machine, pulling the latest into an existing clone, checking the installed version, or removing it. Triggers on "install coco", "update coco", "set up coco", "coco cli", "uninstall coco".
.claude/skills/coco-research-coco-cli/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | -23% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 5% | 0% |
| case-18 | ✗→✓ | ▲ Improved | -37% | 0% |
| case-19 | ✗→✓ | ▲ Improved | -63% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -51% | 0% |
<!-- Wrapper generated with the cli-anything skill. Methodology: HKUDS/CLI-Anything (https://github.com/HKUDS/CLI-Anything) — Apache-2.0. -->
Thin agent wrapper over bin/coco.js (package name cocosuperintelligence). Drives the real CLI; does not reimplement it.
npx cocosuperintelligence is the published path: it clones the pinned release tag and runs install. From a clone, invoke node bin/coco.js. install / update check out the release tag matching package.json (v1.5.0 today), not floating main.
| Command | Purpose | Invocation | |---|---|---| | (default) | clone + install, auto-detecting the adapter | node bin/coco.js | | install | clone + install with explicit flags | node bin/coco.js install [flags] | | update | check out the pinned release tag | node bin/coco.js update [dir] | | uninstall | remove symlinks + the clone | node bin/coco.js uninstall [dir] | | version | print version + check for updates | node bin/coco.js version | | --help | print usage | node bin/coco.js --help |
--adapter <name> — one of claude-code | cursor | grok | codex | generic--systems <list> — comma-separated, e.g. gsd,brain,team--dry-run — preview only, no writesnode bin/coco.js install --adapter cursornode bin/coco.js install --adapter groknode bin/coco.js install --systems gsd,brain --adapter claude-codenode bin/coco.js install --dry-runnode bin/coco.js updateThis CLI is human-output-oriented; it has no native --json mode. For agent use:
version prints cocosuperintelligence vX.Y.Z plus an update hint — parse the vX.Y.Z token.install.sh step failed — surface the captured stdout + stderr to the user.github.com only; on network failure the version check degrades quietly and the command still runs.install clones the pinned release tag and creates symlinks; uninstall removes them. install/update re-fetch, so they are not no-ops. Branch checkouts still git pull --ff-only and are not force-moved onto an older pin.github.com only. No telemetry. Disable update checks with COCO_NO_UPDATE_CHECK=1.node/npx, git, and bash. npx cocosuperintelligence is the published path (clones the pin and installs).| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 16,425 | 16,491 | +0% | 1 | 1 | 0% | 263 | 1,042 | +296% | 0 | 0 | — |
case-02 | fail→fail | 21,173 | 12,680 | -40% | 1 | 1 | 0% | 941 | 889 | -6% | 0 | 0 | — |
case-08 | fail→pass | 13,893 | 6,550 | -53% | 1 | 1 | 0% | 2,209 | 1,708 | -23% | 0 | 0 | — |
case-13 | fail→pass | 12,185 | 3,320 | -73% | 1 | 1 | 0% | 1,080 | 1,139 | +5% | 0 | 0 | — |
case-18 | fail→pass | 17,338 | 2,966 | -83% | 1 | 1 | 0% | 1,702 | 1,068 | -37% | 0 | 0 | — |
case-19 | fail→pass | 24,146 | 7,856 | -67% | 1 | 1 | 0% | 2,870 | 1,069 | -63% | 0 | 0 | — |
case-20 | pass→pass | 16,525 | 13,847 | -16% | 1 | 1 | 0% | 2,126 | 2,186 | +3% | 0 | 0 | — |
case-21 | pass→pass | 13,797 | 4,846 | -65% | 1 | 1 | 0% | 1,351 | 1,460 | +8% | 0 | 0 | — |
case-03 | fail→fail | 13,012 | 17,646 | +36% | 1 | 1 | 0% | 1,497 | 1,001 | -33% | 0 | 0 | — |
case-04 | fail→pass | 16,222 | 2,490 | -85% | 1 | 1 | 0% | 2,021 | 981 | -51% | 0 | 0 | — |
case-05 | fail→pass | 16,792 | 8,298 | -51% | 1 | 1 | 0% | 2,041 | 1,222 | -40% | 0 | 0 | — |
case-06 | fail→pass | 11,568 | 10,139 | -12% | 1 | 1 | 0% | 1,238 | 1,426 | +15% | 0 | 0 | — |
case-07 | fail→pass | 10,977 | 9,629 | -12% | 1 | 1 | 0% | 1,529 | 1,440 | -6% | 0 | 0 | — |
case-09 | fail→pass | 14,824 | 8,510 | -43% | 1 | 1 | 0% | 1,850 | 1,199 | -35% | 0 | 0 | — |
case-10 | fail→pass | 12,152 | 7,186 | -41% | 1 | 1 | 0% | 1,243 | 900 | -28% | 0 | 0 | — |
case-11 | fail→fail | 18,352 | 9,162 | -50% | 1 | 1 | 0% | 2,081 | 1,234 | -41% | 0 | 0 | — |
case-12 | fail→pass | 13,203 | 2,033 | -85% | 1 | 1 | 0% | 713 | 923 | +29% | 0 | 0 | — |
case-14 | pass→pass | 18,514 | 4,609 | -75% | 1 | 1 | 0% | 2,103 | 1,388 | -34% | 0 | 0 | — |
case-15 | fail→fail | 9,172 | 1,363 | -85% | 1 | 1 | 0% | 490 | 820 | +67% | 0 | 0 | — |
case-16 | fail→pass | 11,057 | 10,458 | -5% | 1 | 1 | 0% | 1,765 | 1,536 | -13% | 0 | 0 | — |
case-17 | fail→pass | 19,645 | 11,260 | -43% | 1 | 1 | 0% | 2,221 | 1,850 | -17% | 0 | 0 | — |
case-22 | pass→pass | 12,672 | 4,571 | -64% | 1 | 1 | 0% | 1,377 | 1,425 | +3% | 0 | 0 | — |
case-23 | fail→pass | 12,575 | 7,869 | -37% | 1 | 1 | 0% | 1,339 | 1,134 | -15% | 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 20 counted toward the lift figure. The other 3 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 +61 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.
The publisher has shipped newer versions since this run, so these numbers describe v1, not the version currently listed.
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.