convez.ai
SkillAutomaçãoAvançado
EudOnline

Infinitas Skill

EudOnline/infinitas-skill

Registry e control plane privado para distribuição de skills com API object-centric

Ver no GitHub
InicialPluginScore · de 0 a 100

Registro privado de skills e plano de controle hospedado para gerenciamento de agentes Claude. Sistema production-ready que consolida lógica em um pacote Python mantido, CLI único e documentação estruturada por papéis (contribuidores, operadores, integradores).

Instalação

$npx convez-ai EudOnline/infinitas-skill

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
71%
Último commit
há 1 mês
Versão
—
Issues abertas
0
Contribuidores
4

Casos de uso

  • Gerenciar versões e distribuição de skills em ambiente enterprise
  • Publicar e controlar acesso a agentes e automações via API
  • Administrar tokens, visibilidade e compartilhamento de objetos através de web UI

Veredito

Infraestrutura madura para operações de skills em produção. Requer experiência com Python, APIs REST e operações — não é ferramenta de uso direto, mas backbone técnico para times que automatizam em escala com Claude agents.

README

A tradução para o português deste README ainda está na fila.


audience: contributors, operators, integrators owner: repository maintainers source_of_truth: repo entry page last_reviewed: 2026-06-01 status: maintained

infinitas-skill

Private-first skill registry and hosted control plane.

This repository is production-oriented and in an active maintainability hardening phase. The hosted backend remains private-first, while the maintained agent runtime model now follows OpenClaw-first semantics and the code and docs continue consolidating around one Python package, one maintained CLI, and one role-based documentation tree. The registry/release/install backend remains the durable control plane. Human operators now use the Library web admin surface for distribution work, while agents and automation publish through object-centric APIs.

Start here

Repository shape

  • src/infinitas_skill/: maintained Python package and future home for shared runtime logic
  • scripts/: repository automation, validation, packaging, and compatibility-era helpers that are being actively reduced
  • server/: hosted control-plane runtime
  • docs/: role-based documentation during the reset

Maintained surfaces

  • package-owned: src/infinitas_skill/install/..., src/infinitas_skill/policy/..., src/infinitas_skill/release/..., and src/infinitas_skill/server/... own maintained CLI logic.
  • runtime-owned: server/modules/... and server/ui/... own hosted API, UI, and presentation logic, while server/app.py stays focused on app assembly.
  • automation-owned: remaining top-level scripts exist for repository automation, catalog generation, release packaging, and targeted regression coverage; canonical user-facing command surfaces live under infinitas.

Hosted web admin

The maintained browser product is the distribution admin, not the legacy authoring console. Primary routes:

  • /manage — consolidated admin console (Library, Access, Shares, Activity)
  • /library/{object_id} — object detail
  • /library/{object_id}/releases/{release_id} — release detail
  • /settings

The web UI uses product vocabulary: Object, Release, Visibility, Token, Share Link, and Activity. Old /library, /access, /shares, and /activity routes redirect to /manage. Legacy /skills, /drafts, /releases/{id}, /access/tokens, and /review-cases UI routes also redirect to /manage. Low-level authoring routes remain available for API and CLI workflows.

Human operators authenticate through the web UI with username + password (session cookie). Agents and automation use Bearer tokens via the API.

Agent and automation workflows use these product APIs:

  • GET /api/library, GET /api/library/{object_id}, GET /api/library/{object_id}/releases
  • PUT /api/publish/objects/{slug}
  • POST /api/publish/objects/{object_id}/releases
  • GET /api/publish/releases/{release_id}/status
  • POST /api/objects/{object_id}/tokens, GET /api/objects/{object_id}/tokens
  • POST /api/tokens/{token_id}/revoke
  • POST /api/releases/{release_id}/share-links, GET /api/releases/{release_id}/share-links
  • POST /api/share-links/{share_id}/resolve, POST /api/share-links/{share_id}/revoke
  • GET /api/activity, GET /api/tokens/{token_id}/activity, GET /api/share-links/{share_id}/activity

Net-new web admin templates are generated or iterated from the Kimi CLI spec in docs/specs/kimi-library-ui-spec.md; hand edits should stay limited to integration glue, route wiring, and bug fixes.

Maintained CLI surface

Maintained entrypoints introduced so far:

uv run infinitas compatibility check-platform-contracts --max-age-days 30 --stale-policy fail
uv run infinitas install resolve-plan --skill-dir templates/basic-skill --target-dir .tmp-installed-skills --json
uv run infinitas policy check-packs
uv run infinitas policy recommend-reviewers <skill> --as-active --json
uv run infinitas policy review-status <skill> --as-active --require-pass --json
uv run infinitas registry --help
uv run infinitas release check-state <skill> --mode local-preflight --json
uv run infinitas release signing-readiness --skill <skill> --json
uv run infinitas release doctor-signing <skill> --json
uv run infinitas release bootstrap-signing --help
uv run infinitas server healthcheck --api-url http://127.0.0.1:8000 --repo-path /srv/infinitas/repo --artifact-path /srv/infinitas/artifacts --database-url sqlite:////srv/infinitas/data/server.db --json
uv run infinitas server memory-curation --database-url sqlite:////srv/infinitas/data/server.db --action plan --limit 50 --json
uv run infinitas server memory-curation --database-url sqlite:////srv/infinitas/data/server.db --action archive --apply --max-actions 10 --json
uv run infinitas server memory-curation --database-url sqlite:////srv/infinitas/data/server.db --action prune --apply --max-actions 5 --json
uv run infinitas server memory-curation --database-url sqlite:////srv/infinitas/data/server.db --action archive --apply --max-actions 10 --enqueue --json
uv run infinitas server memory-curation --database-url sqlite:////srv/infinitas/data/server.db --use-server-policy --enqueue --json
uv run infinitas server memory-health --database-url sqlite:////srv/infinitas/data/server.db --limit 20 --json
uv run infinitas server memory-observability --database-url sqlite:////srv/infinitas/data/server.db --limit 20 --job-limit 10 --window-hours 24 --json
uv run infinitas server memory-baselines --database-url sqlite:////srv/infinitas/data/server.db --window-hours 24 --json
uv run infinitas server prune-backups --backup-root /srv/infinitas/backups --keep-last 7 --json

uv run infinitas ... is the maintained CLI surface. New command surfaces should land under infinitas, not as new top-level scripts.

Local verification

Preferred maintained-surface entrypoints:

make bootstrap
make clean-local
make ci-fast
make test-fast
make test-full
make lint-maintained
uv run pytest tests/integration/test_memory_evaluation_matrix.py -q

Treat verification as three layers:

  • maintained runtime and backend truth: make lint-maintained, make test-fast
  • migration-focused OpenClaw runtime matrix: uv run pytest tests/unit/openclaw tests/integration/test_openclaw_runtime_index.py tests/integration/test_openclaw_install_planning.py tests/integration/test_cli_openclaw_runtime.py tests/integration/test_cli_release_state.py tests/integration/test_memory_evaluation_matrix.py -q
  • broader repository sweeps: ./scripts/check-all.sh focused-integration, then opt into release-long when release packaging or attestation flows changed

make clean-local is the supported local hygiene path for generated artifacts and local automation output (__pycache__, *.pyc, test/lint caches, local *.egg-info metadata, and output/playwright). It intentionally preserves tracked files, including placeholders such as build/.gitkeep.

make ci-fast mirrors the maintained CI fast gate by running make lint-maintained and then make test-fast before the repository drops to the broader closeout matrix.

make test-fast is now the default fast path for maintained work. It covers the focused integration tier, the promoted high-value pytest regressions, and the maintainability budget gate before you drop to raw fallback commands.

Raw commands remain available as fallback detail:

uv sync
uv run pytest tests/integration/test_cli_release_state.py tests/integration/test_cli_server_ops.py tests/integration/test_private_registry_ui.py -q
uv run pytest tests/integration/test_cli_install_planning.py -q
uv run pytest tests/integration/test_cli_policy.py -q
uv run ruff check src/infinitas_skill server/ui server/app.py tests/integration tests/unit
uv run python3 scripts/test-platform-contracts.py
uv run python3 scripts/test-install-manifest-compat.py
uv run python3 scripts/test-release-invariants.py
uv run python3 scripts/test-infinitas-cli-platform-contracts.py
uv run python3 scripts/test-infinitas-cli-registry.py
uv run python3 scripts/test-infinitas-cli-server.py
uv run python3 scripts/test-infinitas-cli-reference-docs.py
uv run python3 scripts/test-doc-governance.py
./scripts/check-all.sh
./scripts/check-all.sh release-long

Use ./scripts/check-all.sh release-long as the canonical opt-in long-running pre-release gate. It is the smallest named block that proves the transparency-log and release-invariant flows end to end.

Legacy compatibility coverage still matters, but it is now regression and audit coverage rather than the maintained runtime gate. Release blocking follows canonical OpenClaw runtime freshness.

make lint-maintained currently enforces the maintained-surface E/F/I baseline while temporarily deferring E501 only in the current debt-heavy maintained files, plus a few legacy path-bootstrap E402 cases, until the planned module splits land.

Hard maintainability budgets now backstop the maintained reset:

  • server/app.py must stay at or below 80 lines
  • src/infinitas_skill/server/ops.py must stay at or below 550 lines
  • src/infinitas_skill/install/service.py must stay at or below 650 lines
  • src/infinitas_skill/release/service.py must stay at or below 650 lines
  • top-level files under scripts/ must stay at or below 231 until a deliberate cleanup changes the ceiling

tests/integration/test_maintainability_budgets.py and scripts/check-all.sh focused-integration enforce these limits.

Local runs default to INFINITAS_SERVER_ENV=development. Use INFINITAS_SERVER_ENV=test when you need fixture-safe automated behavior.

Maintainability reset rules

  • No new top-level script may be added under scripts/ without explicit architecture approval.
  • Do not raise maintained-module line budgets or the top-level script ceiling without updating docs and the budget test in the same change.
  • No new long-lived doc may be added outside docs/guide/, docs/reference/, docs/ops/, docs/archive/, or docs/adr/.
  • New shared Python logic should land under src/infinitas_skill/.
  • Do not add or restore compatibility wrapper entrypoints once a canonical infinitas ... command exists.

Product and policy context

OpenClaw is now the canonical agent runtime.

The internal backend lifecycle remains:

skill -> draft -> sealed version -> release -> exposure -> review case -> grant/credential -> discovery/install

That lifecycle is an implementation detail for the web product. Human admins should see the Library distribution model, and agents should use the object-centric publish API instead of manually driving draft/seal routes for the happy path.

Use these canonical docs for the current model:

Compatibility reporting still distinguishes between legacy declared support metadata such as _meta.json.agent_compatible and historical verified support evidence. Those fields remain useful for migration and audit context, but they are no longer the maintained runtime source of truth.

The optional memory layer is additive only:

  • recommendation and inspect may include advisory memory fields
  • recommendation and inspect can now emit local memory_retrieval audit events when INFINITAS_DISCOVERY_AUDIT_DATABASE_URL is configured
  • lifecycle events may emit best-effort memory writeback attempts plus traceable audit events
  • fixture-backed memory evaluation now lives in tests/integration/test_memory_evaluation_matrix.py
  • the evaluation matrix now checks duplicate suppression and noisy recall stability
  • the evaluation matrix now also publishes deterministic usefulness summary metrics for helpful use vs correct restraint
  • operator-facing curation planning, guarded archive/prune execution, queue-driven enqueue, and server-policy scheduled runs now live behind uv run infinitas server memory-curation ...
  • operator-facing writeback diagnostics now live behind uv run infinitas server memory-health ...
  • operator-facing memory operations summary plus rolling drift context now lives behind uv run infinitas server memory-observability ..., including retrieval-side memory_retrieval status and effect visibility
  • operator-facing rolling memory baselines now live behind uv run infinitas server memory-baselines ..., including retrieval-side drift
  • release, review, access, and install truth still comes from the local database, immutable artifacts, and current policy checks

Policy trace and validation output

Policy-aware commands continue to expose structured diagnostics for operators and automation:

  • uv run infinitas policy check-promotion <skill> --json returns a policy_trace payload for promotion decisions.
  • uv run infinitas policy review-status <skill> --as-active --show-recommendations --json returns review gate status plus reviewer guidance.
  • uv run infinitas release signing-readiness --skill operate-infinitas-skill --json summarizes repo signing trust and per-skill readiness.
  • uv run infinitas release check-state operate-infinitas-skill --json returns the release decision plus policy_trace details.
  • scripts/validate-registry.py --json returns validation_errors alongside namespace-level policy_trace data.
  • policy/policy-packs.json selects ordered shared defaults from policy/packs/*.json, while repository-local policy files remain the last override layer.
  • policy/team-policy.json remains the default team-governance input that keeps review and release checks aligned.
Ver o original no GitHub

Bom para

Relacionados