Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Build, repair, or package Agentlas agents and teams with Hephaestus.
.claude/skills/agentlas-ai-hep-build/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 50% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 140% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 240% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 17% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 115% | 0% |
Update fallback: 자동 업데이트가 안 되면 hephaestus update를 한 번 실행하세요. 업데이트하지 않아도 현재 버전 명령은 그대로 동작합니다.
Raw arguments: everything the user typed after /skill:hep-build.
Use the agentlas-core-engine-meta-agent skill from the Hephaestus plugin. Expose /skill:hep-build as the public Kimi Code CLI skill next to /skill:hep-network and /skill:hep-cloud.
Resolve the installed engine first:
bashENGINE="" for candidate in \ "${KIMI_PLUGIN_ROOT:-}" "${CLAUDE_PLUGIN_ROOT:-}" "${PLUGIN_ROOT:-}" \ "$HOME/.agentlas/runtime/current/host_adapters/kimi/skills/hephaestus-build" \ "$HOME/.agentlas/runtime/current/host_adapters/codex/plugins/agentlas-core-engine-meta-agent" \ "$HOME/.agentlas/runtime/current/host_adapters/claude/plugins/agentlas-core-engine-meta-agent" \ "$HOME/.agentlas/runtime/current" "." do if [ -n "$candidate" ] && [ -f "$candidate/AGENTS.md" ] && [ -f "$candidate/package-contract.json" ] && [ -f "$candidate/contracts/builder-interview-research-gate.md" ]; then ENGINE="$candidate"; break fi done [ -n "$ENGINE" ] || { echo "Hephaestus engine not found. Run the installer first." >&2; exit 1; } RUNNER="$HOME/.agentlas/runtime/current/bin/hephaestus" [ -x "$RUNNER" ] || RUNNER="$ENGINE/bin/hephaestus" [ -x "$RUNNER" ] || { echo "Hephaestus runner not found under $ENGINE." >&2; exit 1; }
Read contracts only from $ENGINE. Take exactly one user-named or confirmed folder as PACKAGE_TARGET; if none or multiple candidates exist, stop and ask. Never default to ., cwd, or $ENGINE. Run "$RUNNER" contract resolve-target "$PACKAGE_TARGET" --base "$PWD" and set PACKAGE_ROOT only to the status-ok receipt's exact package_root before any package command.
ontology, resolve the runner exactly as in/skill:hep-network and run "$RUNNER" ontology.
multi-agent-team-builder, or agentlas-packager by independent ownership boundaries and execute the meta-agent procedure on: $ARGUMENTS. If single↔multi is unclear, ask first in plain language: "이 일을 한 명의 전문가가 처음부터 끝까지 맡으면 되나요, 아니면 조사/분석/검토처럼 여러 전문가가 나눠 맡고 마지막에 합쳐야 하나요?" Do not expose internal labels like ownership boundary, memory/context, synthesis, or produces/consumes.
Research Gate from $ENGINE/contracts/builder-interview-research-gate.md. Follow the briefing interview engine (agentlas_cloud/interview/) and write .agentlas/work-brief.json (work-brief/1.0). Ask an 8-12 question first batch when the request is vague, continue follow-ups until the functional brief is clear, research official sources, similar agent repositories or comparables, academic/professional theory, and plugin docs, compare selected and rejected tools/plugins, synthesize domain-expert behavior, and create docs/builder-interview.md, docs/research-sources.md, docs/tool-selection.md, docs/domain-expert-synthesis.md, docs/prompt-performance-contract.md, and .agentlas/capability-eval-plan.json.
AGENTS.md, CLAUDE.md, GEMINI.md, agent.md, skills, workflow/command adapters, runtime prompts, handoff contracts, return contracts, and operating docs. Translate Korean or other-language source material into English agent behavior. Localized marketplace copy, routing trigger examples, and sample user inputs may use the target user language.
Before writing any package file, lay the contract down: "$RUNNER" contract scaffold "$PACKAGE_ROOT" --mode single|team|package. Then, as soon as the routing card exists, run "$RUNNER" contract complete "$PACKAGE_ROOT" --mode single|team|package — the engine fills every artifact the package already answers (agent.md, work brief, sitemap, routing benchmarks, capability eval plan, builder interview, research sources, output example) from the routing card, the roster, and the schemas on disk. It never overwrites an authored body and never invents a fact. Run it BEFORE contract verify, so verify reports only the genuinely authored half. It copies every required artifact into place with named {{PLACEHOLDER}} holes and never overwrites. Skipping it is how a build ends with 5 of 18 required artifacts and still reports success. contract prompt --mode <mode> lists what each one is for.
"$RUNNER" contract verify "$PACKAGE_ROOT" --mode single|team|package (this runs the team-shape rule too). If it fails, do not report completed; correct the shape by collapsing to a valid single-agent package or adding orchestrator/HQ plus company-blueprint topology.
global_commands for the created agent or team in the finalresponse, plus interview_research evidence.
local discovery immediately: run "$RUNNER" cards migrate "$PACKAGE_ROOT" --tier local --overwrite (or the same hephaestus runner in cache if local binary is unavailable), and include migration result in evidence.
If runtime discovery migration isn't needed, still validate that the package has ./.agentlas/routing-card.json and include that local-card artifact in evidence when skipping migration.
After the package is verified and saved locally, use the host's structured choice UI when available and ask exactly one final question:
same account's other Desktops. Mobile can use it only through a paired Desktop after that Desktop restores/installs it; this is not hosted model execution.
network mutation.
Do not upload automatically. A missing answer or non-interactive host defaults to local-only. On explicit Cloud consent, run the resolved Hephaestus runner as "$RUNNER" upload "$PACKAGE_ROOT" --visibility private-link, where PACKAGE_ROOT is the exact verified package root, never a guessed parent. Cloud auth/offline/CAS/security failure must not delete or roll back the local package; report the failure and exact retry command. Public Hub publication is a separate explicit action and must not appear as a third choice here.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 61,768 | 14,276 | -77% | 1 | 1 | 0% | 10,585 | 2,510 | -76% | 0 | 0 | — |
case-02 | fail→fail | 29,968 | 11,058 | -63% | 1 | 1 | 0% | 5,640 | 2,251 | -60% | 0 | 0 | — |
case-03 | fail→fail | 28,810 | 12,936 | -55% | 1 | 1 | 0% | 1,656 | 2,156 | +30% | 0 | 0 | — |
case-04 | fail→fail | 13,807 | 14,774 | +7% | 1 | 1 | 0% | 1,822 | 2,676 | +47% | 0 | 0 | — |
case-05 | fail→fail | 5,493 | 14,138 | +157% | 1 | 1 | 0% | 700 | 2,333 | +233% | 0 | 0 | — |
case-06 | fail→fail | 11,398 | 13,330 | +17% | 1 | 1 | 0% | 1,178 | 2,498 | +112% | 0 | 0 | — |
case-07 | fail→pass | 10,350 | 5,295 | -49% | 1 | 1 | 0% | 1,618 | 2,427 | +50% | 0 | 0 | — |
case-08 | fail→pass | 7,020 | 17,253 | +146% | 1 | 1 | 0% | 949 | 2,279 | +140% | 0 | 0 | — |
case-09 | fail→fail | 12,432 | 26,815 | +116% | 1 | 1 | 0% | 1,642 | 2,719 | +66% | 0 | 0 | — |
case-10 | pass→fail | 10,377 | 31,563 | +204% | 1 | 1 | 0% | 1,409 | 2,949 | +109% | 0 | 0 | — |
case-11 | fail→fail | 9,790 | 17,276 | +76% | 1 | 1 | 0% | 1,343 | 2,714 | +102% | 0 | 0 | — |
case-12 | fail→fail | 6,959 | 14,851 | +113% | 1 | 1 | 0% | 909 | 2,562 | +182% | 0 | 0 | — |
case-13 | fail→pass | 5,716 | 4,816 | -16% | 1 | 1 | 0% | 704 | 2,397 | +240% | 0 | 0 | — |
case-14 | fail→pass | 14,057 | 5,435 | -61% | 1 | 1 | 0% | 2,180 | 2,542 | +17% | 0 | 0 | — |
case-15 | fail→fail | 9,617 | 16,294 | +69% | 1 | 1 | 0% | 1,259 | 2,942 | +134% | 0 | 0 | — |
case-16 | fail→pass | 8,660 | 6,763 | -22% | 1 | 1 | 0% | 1,159 | 2,495 | +115% | 0 | 0 | — |
case-17 | pass→pass | 9,820 | 2,558 | -74% | 1 | 1 | 0% | 1,384 | 1,962 | +42% | 0 | 0 | — |
case-18 | fail→pass | 13,342 | 5,388 | -60% | 1 | 1 | 0% | 1,937 | 2,495 | +29% | 0 | 0 | — |
case-19 | fail→fail | 26,813 | 13,375 | -50% | 1 | 1 | 0% | 3,184 | 2,572 | -19% | 0 | 0 | — |
case-20 | fail→pass | 7,573 | 8,916 | +18% | 1 | 1 | 0% | 870 | 2,806 | +223% | 0 | 0 | — |
case-21 | fail→pass | 10,046 | 11,636 | +16% | 1 | 1 | 0% | 1,347 | 2,390 | +77% | 0 | 0 | — |
case-22 | pass→pass | 8,738 | 4,627 | -47% | 1 | 1 | 0% | 1,105 | 2,218 | +101% | 0 | 0 | — |
case-23 | fail→pass | 48,388 | 5,529 | -89% | 1 | 1 | 0% | 3,690 | 2,459 | -33% | 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 10 counted toward the lift figure. The other 13 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 +35 percentage points is the difference between those two pass rates over the 10 comparable cases. 5 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.