Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create, update, version, and manage Observal agents. Use when the user wants to create a new agent, update an existing one, release a new version, scaffold a YAML project, add components, build, publish, bulk-create, archive, delete, or restore agents.
.claude/skills/observal-observal-agents/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -10% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -39% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -55% | 0% |
--output json whenever supported. Parse list results from items and pagination fields.--help before acting when a path or flag is uncertain.--no-prompt, and confirmation flags.qualified_name values returned by JSON. Never automate with row numbers.agent init, agent add, and agent build over hand-written scaffolding or custom validation.| User intent | Workflow | | --- | --- | | Find or inspect an Agent | Discover and inspect | | Install an existing Agent into a harness | Pull and verify | | Create a simple Agent in one call | Direct create | | Author an Agent with components or files | Init, add, build, publish | | Change the current listing without a new reviewed version | Update in place | | Publish a reviewed patch, minor, or major version | Release | | Create many Agents from a prepared file | Bulk create | | Archive, restore, transfer, or manage co-authors | Lifecycle and collaboration |
Read Agent workflows completely before executing the selected workflow.
create without complete flags starts a wizard. Agents must provide all required inputs or use a file.publish --update changes the current Agent in place.release --bump creates a reviewed version. Do not use update when the user asked for a release.files, warnings, and setup_commands. Partial setup is not success.Report the canonical Agent identity, resulting status and version, files changed for local operations, warnings, and the smallest next action. Verify with agent show, agent versions, or scan when the mutation response is not sufficient.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | fail→pass | 10,929 | 7,765 | -29% | 1 | 1 | 0% | 1,765 | 1,597 | -10% | 0 | 0 | — |
case-01 | fail→fail | 4,162 | 5,725 | +38% | 1 | 1 | 0% | 255 | 805 | +216% | 0 | 0 | — |
case-02 | fail→fail | 6,415 | 7,619 | +19% | 1 | 1 | 0% | 1,110 | 771 | -31% | 0 | 0 | — |
case-03 | fail→fail | 5,185 | 5,714 | +10% | 1 | 1 | 0% | 260 | 877 | +237% | 0 | 0 | — |
case-04 | pass→pass | 6,913 | 5,054 | -27% | 1 | 1 | 0% | 1,203 | 1,338 | +11% | 0 | 0 | — |
case-05 | pass→pass | 7,379 | 4,767 | -35% | 1 | 1 | 0% | 1,244 | 1,401 | +13% | 0 | 0 | — |
case-06 | pass→fail | 8,148 | 10,228 | +26% | 1 | 1 | 0% | 1,418 | 2,095 | +48% | 0 | 0 | — |
case-08 | fail→pass | 10,175 | 2,452 | -76% | 1 | 1 | 0% | 1,593 | 889 | -44% | 0 | 0 | — |
case-09 | pass→pass | 8,977 | 3,759 | -58% | 1 | 1 | 0% | 1,379 | 882 | -36% | 0 | 0 | — |
case-10 | fail→pass | 12,163 | 5,484 | -55% | 1 | 1 | 0% | 1,778 | 1,393 | -22% | 0 | 0 | — |
case-11 | pass→pass | 10,162 | 3,376 | -67% | 1 | 1 | 0% | 1,370 | 1,060 | -23% | 0 | 0 | — |
case-12 | pass→pass | 5,340 | 4,665 | -13% | 1 | 1 | 0% | 845 | 1,318 | +56% | 0 | 0 | — |
case-13 | fail→pass | 12,095 | 3,825 | -68% | 1 | 1 | 0% | 1,819 | 1,105 | -39% | 0 | 0 | — |
case-14 | fail→pass | 14,133 | 2,847 | -80% | 1 | 1 | 0% | 2,014 | 912 | -55% | 0 | 0 | — |
case-15 | pass→pass | 13,252 | 3,987 | -70% | 1 | 1 | 0% | 1,946 | 1,088 | -44% | 0 | 0 | — |
case-16 | pass→pass | 10,205 | 5,831 | -43% | 1 | 1 | 0% | 1,431 | 1,429 | -0% | 0 | 0 | — |
case-17 | fail→pass | 11,668 | 5,286 | -55% | 1 | 1 | 0% | 1,855 | 1,349 | -27% | 0 | 0 | — |
case-18 | pass→pass | 33,091 | 7,449 | -77% | 1 | 1 | 0% | 2,583 | 1,786 | -31% | 0 | 0 | — |
case-19 | pass→fail | 6,721 | 2,228 | -67% | 1 | 1 | 0% | 1,018 | 769 | -24% | 0 | 0 | — |
case-20 | fail→fail | 10,978 | 2,461 | -78% | 1 | 1 | 0% | 1,603 | 945 | -41% | 0 | 0 | — |
case-21 | fail→pass | 13,233 | 6,584 | -50% | 1 | 1 | 0% | 2,081 | 1,575 | -24% | 0 | 0 | — |
case-22 | pass→pass | 4,539 | 3,656 | -19% | 1 | 1 | 0% | 645 | 1,110 | +72% | 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 19 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 +23 percentage points is the difference between those two pass rates over the 19 comparable cases. 3 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/13/2026 | +35% |
Other measured skills in the registry, with their headline benchmark lift.