Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Install workspace skills from curated sources or import open-source skills from GitHub.
.claude/skills/holaboss-ai-skill-installer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | 81% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -47% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -12% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -57% | 0% |
Use this skill to install workspace skills into the workspace-local skills/ directory.
A freshly installed skill is usable immediately, in this same run: skill({ name: "<id>" }) reads the folder off disk on every call. What waits for the next run is only the skill listing in your capability manifest — the skill itself is already there. Never tell the user to start a new conversation to use what you just installed; if they asked you to install it and use it, load it and carry on.
Open-source skills (e.g. github.com/anthropics/skills) are just SKILL.md folders in the same format used here, so they can be imported whole. Use the runtime endpoint — it fetches the entire folder (SKILL.md plus bundled scripts/, references/, assets/), maps foreign frontmatter (allowed-tools → holaboss_granted_tools, aligns name to the installed id), and writes it under skills/<id>/:
POST /api/v1/workspaces/{workspaceId}/skills/import-github/preview with { "url": "<github folder or SKILL.md URL>" } — returns the parsed name, description, granted tools, and file list so you can confirm before installing.POST /api/v1/workspaces/{workspaceId}/skills/import-github with the same body.Accepts github.com/<owner>/<repo>/tree/<ref>/<path> (folder), .../blob/<ref>/<path>/SKILL.md (single file), or a bare repo. Pass an optional "ref" for a specific branch/tag/commit (defaults to the repo's default branch). Skills whose bundled scripts assume tools this sandbox lacks still install and work as pure guidance.
Its name and how to use it — the id they invoke it by, the one-line description, and what to say to put it to work. The install response also carries the written file list; that is for your own verification, not something to report. "Installed SKILL.md, scripts/render.py" tells them nothing they can act on.
skills/<skill-id>/ with its SKILL.md plus any helper files. For skills you author by hand, create these files directly with the Write tool.runtime/harnesses/src/embedded-skills/. Do not install into $CODEX_HOME/skills unless the user explicitly asks for a global install rather than a workspace install.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 8,357 | 20,287 | +143% | 1 | 1 | 0% | 555 | 1,194 | +115% | 0 | 0 | — |
case-02 | fail→fail | 9,645 | 14,241 | +48% | 1 | 1 | 0% | 1,692 | 1,221 | -28% | 0 | 0 | — |
case-03 | fail→fail | 8,536 | 16,434 | +93% | 1 | 1 | 0% | 1,220 | 896 | -27% | 0 | 0 | — |
case-04 | fail→fail | 8,498 | 12,821 | +51% | 1 | 1 | 0% | 1,488 | 1,247 | -16% | 0 | 0 | — |
case-05 | fail→fail | 9,042 | 14,094 | +56% | 1 | 1 | 0% | 1,991 | 1,121 | -44% | 0 | 0 | — |
case-06 | fail→pass | 4,484 | 3,798 | -15% | 1 | 1 | 0% | 627 | 1,136 | +81% | 0 | 0 | — |
case-07 | pass→pass | 8,462 | 4,252 | -50% | 1 | 1 | 0% | 1,595 | 1,126 | -29% | 0 | 0 | — |
case-08 | pass→pass | 7,151 | 3,500 | -51% | 1 | 1 | 0% | 1,459 | 1,179 | -19% | 0 | 0 | — |
case-09 | pass→pass | 6,858 | 1,974 | -71% | 1 | 1 | 0% | 1,063 | 954 | -10% | 0 | 0 | — |
case-10 | fail→pass | 11,975 | 4,967 | -59% | 1 | 1 | 0% | 2,329 | 1,241 | -47% | 0 | 0 | — |
case-11 | pass→fail | 12,380 | 18,361 | +48% | 1 | 1 | 0% | 1,939 | 1,174 | -39% | 0 | 0 | — |
case-12 | fail→fail | 13,897 | 8,714 | -37% | 1 | 1 | 0% | 1,998 | 2,032 | +2% | 0 | 0 | — |
case-13 | fail→pass | 9,984 | 5,065 | -49% | 1 | 1 | 0% | 1,824 | 1,606 | -12% | 0 | 0 | — |
case-14 | pass→pass | 10,552 | 3,724 | -65% | 1 | 1 | 0% | 1,837 | 1,261 | -31% | 0 | 0 | — |
case-15 | fail→pass | 12,213 | 2,876 | -76% | 1 | 1 | 0% | 1,672 | 1,080 | -35% | 0 | 0 | — |
case-16 | fail→pass | 15,274 | 2,478 | -84% | 1 | 1 | 0% | 2,120 | 913 | -57% | 0 | 0 | — |
case-17 | pass→pass | 13,281 | 2,717 | -80% | 1 | 1 | 0% | 2,216 | 1,032 | -53% | 0 | 0 | — |
case-18 | fail→pass | 11,801 | 4,262 | -64% | 1 | 1 | 0% | 1,574 | 1,138 | -28% | 0 | 0 | — |
case-19 | fail→pass | 13,854 | 3,602 | -74% | 1 | 1 | 0% | 2,196 | 1,159 | -47% | 0 | 0 | — |
case-20 | pass→fail | 5,036 | 7,426 | +47% | 1 | 1 | 0% | 824 | 817 | -1% | 0 | 0 | — |
case-21 | pass→fail | 11,354 | 12,374 | +9% | 1 | 1 | 0% | 1,503 | 1,387 | -8% | 0 | 0 | — |
case-22 | pass→fail | 4,341 | 10,743 | +147% | 1 | 1 | 0% | 537 | 1,230 | +129% | 0 | 0 | — |
case-23 | pass→fail | 8,557 | 8,754 | +2% | 1 | 1 | 0% | 722 | 1,078 | +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. 23 cases were attempted, and 14 counted toward the lift figure. The other 9 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 +9 percentage points is the difference between those two pass rates over the 14 comparable cases. 7 cases got worse with the skill loaded, and they are 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.