agent-devkit 0.1.0 → 0.1.6
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +42 -2
- package/package.json +1 -1
- package/runtime/README.md +119 -5
- package/runtime/agents/README.md +8 -0
- package/runtime/agents/execution-reviewer/AGENTS.md +8 -0
- package/runtime/agents/execution-reviewer/README.md +13 -0
- package/runtime/agents/execution-reviewer/agent.yaml +25 -0
- package/runtime/agents/execution-reviewer/capabilities/review-agent-result/capability.yaml +15 -0
- package/runtime/agents/execution-reviewer/capabilities/review-agent-result/decision-rules.md +6 -0
- package/runtime/agents/execution-reviewer/capabilities/review-agent-result/workflow.md +7 -0
- package/runtime/agents/execution-reviewer/capabilities/review-final-output/capability.yaml +15 -0
- package/runtime/agents/execution-reviewer/capabilities/review-final-output/decision-rules.md +6 -0
- package/runtime/agents/execution-reviewer/capabilities/review-final-output/workflow.md +7 -0
- package/runtime/agents/execution-reviewer/capabilities/review-plan/capability.yaml +15 -0
- package/runtime/agents/execution-reviewer/capabilities/review-plan/decision-rules.md +6 -0
- package/runtime/agents/execution-reviewer/capabilities/review-plan/workflow.md +7 -0
- package/runtime/agents/execution-reviewer/infra/README.md +4 -0
- package/runtime/agents/execution-reviewer/knowledge/context.md +6 -0
- package/runtime/agents/execution-reviewer/knowledge/system.md +18 -0
- package/runtime/agents/execution-reviewer/templates/README.md +4 -0
- package/runtime/agents/local-llm-operator/AGENTS.md +8 -0
- package/runtime/agents/local-llm-operator/README.md +12 -0
- package/runtime/agents/local-llm-operator/agent.yaml +26 -0
- package/runtime/agents/local-llm-operator/capabilities/delegate-operational-task/capability.yaml +15 -0
- package/runtime/agents/local-llm-operator/capabilities/delegate-operational-task/decision-rules.md +6 -0
- package/runtime/agents/local-llm-operator/capabilities/delegate-operational-task/workflow.md +7 -0
- package/runtime/agents/local-llm-operator/capabilities/inspect-local-models/capability.yaml +15 -0
- package/runtime/agents/local-llm-operator/capabilities/inspect-local-models/decision-rules.md +5 -0
- package/runtime/agents/local-llm-operator/capabilities/inspect-local-models/workflow.md +6 -0
- package/runtime/agents/local-llm-operator/capabilities/select-local-worker/capability.yaml +15 -0
- package/runtime/agents/local-llm-operator/capabilities/select-local-worker/decision-rules.md +6 -0
- package/runtime/agents/local-llm-operator/capabilities/select-local-worker/workflow.md +7 -0
- package/runtime/agents/local-llm-operator/infra/README.md +4 -0
- package/runtime/agents/local-llm-operator/knowledge/context.md +6 -0
- package/runtime/agents/local-llm-operator/knowledge/system.md +18 -0
- package/runtime/agents/local-llm-operator/templates/README.md +4 -0
- package/runtime/agents/provider-configurator/AGENTS.md +8 -0
- package/runtime/agents/provider-configurator/README.md +5 -0
- package/runtime/agents/provider-configurator/agent.yaml +25 -0
- package/runtime/agents/provider-configurator/capabilities/collect-provider-credentials/capability.yaml +15 -0
- package/runtime/agents/provider-configurator/capabilities/collect-provider-credentials/decision-rules.md +6 -0
- package/runtime/agents/provider-configurator/capabilities/collect-provider-credentials/workflow.md +7 -0
- package/runtime/agents/provider-configurator/capabilities/configure-provider-source/capability.yaml +15 -0
- package/runtime/agents/provider-configurator/capabilities/configure-provider-source/decision-rules.md +6 -0
- package/runtime/agents/provider-configurator/capabilities/configure-provider-source/workflow.md +8 -0
- package/runtime/agents/provider-configurator/capabilities/validate-provider-readiness/capability.yaml +15 -0
- package/runtime/agents/provider-configurator/capabilities/validate-provider-readiness/decision-rules.md +6 -0
- package/runtime/agents/provider-configurator/capabilities/validate-provider-readiness/workflow.md +7 -0
- package/runtime/agents/provider-configurator/infra/README.md +5 -0
- package/runtime/agents/provider-configurator/knowledge/context.md +8 -0
- package/runtime/agents/provider-configurator/knowledge/system.md +19 -0
- package/runtime/agents/provider-configurator/templates/README.md +4 -0
- package/runtime/agents/task-orchestrator/AGENTS.md +8 -0
- package/runtime/agents/task-orchestrator/README.md +7 -0
- package/runtime/agents/task-orchestrator/agent.yaml +31 -0
- package/runtime/agents/task-orchestrator/capabilities/coordinate-execution/capability.yaml +15 -0
- package/runtime/agents/task-orchestrator/capabilities/coordinate-execution/decision-rules.md +6 -0
- package/runtime/agents/task-orchestrator/capabilities/coordinate-execution/workflow.md +7 -0
- package/runtime/agents/task-orchestrator/capabilities/plan-task/capability.yaml +15 -0
- package/runtime/agents/task-orchestrator/capabilities/plan-task/decision-rules.md +6 -0
- package/runtime/agents/task-orchestrator/capabilities/plan-task/workflow.md +8 -0
- package/runtime/agents/task-orchestrator/capabilities/select-specialists/capability.yaml +15 -0
- package/runtime/agents/task-orchestrator/capabilities/select-specialists/decision-rules.md +6 -0
- package/runtime/agents/task-orchestrator/capabilities/select-specialists/workflow.md +7 -0
- package/runtime/agents/task-orchestrator/infra/README.md +5 -0
- package/runtime/agents/task-orchestrator/knowledge/context.md +9 -0
- package/runtime/agents/task-orchestrator/knowledge/system.md +20 -0
- package/runtime/agents/task-orchestrator/templates/README.md +4 -0
- package/runtime/cli/README.md +111 -10
- package/runtime/cli/aikit/__init__.py +1 -1
- package/runtime/cli/aikit/agent_executor.py +66 -0
- package/runtime/cli/aikit/agent_registry.py +80 -0
- package/runtime/cli/aikit/app_home.py +87 -2
- package/runtime/cli/aikit/audit.py +44 -0
- package/runtime/cli/aikit/configuration_orchestrator.py +291 -0
- package/runtime/cli/aikit/control_router.py +298 -0
- package/runtime/cli/aikit/decision_store.py +175 -0
- package/runtime/cli/aikit/diagnostics.py +2 -1
- package/runtime/cli/aikit/execution_reviewer.py +174 -0
- package/runtime/cli/aikit/fallback.py +48 -2
- package/runtime/cli/aikit/install.py +4 -3
- package/runtime/cli/aikit/llm.py +9 -0
- package/runtime/cli/aikit/local_llm_operator.py +112 -0
- package/runtime/cli/aikit/lock.py +4 -1
- package/runtime/cli/aikit/main.py +604 -21
- package/runtime/cli/aikit/model_router.py +79 -0
- package/runtime/cli/aikit/ollama.py +237 -0
- package/runtime/cli/aikit/orchestrator.py +535 -0
- package/runtime/cli/aikit/provider_wizard.py +19 -0
- package/runtime/cli/aikit/review_gate.py +40 -0
- package/runtime/cli/aikit/sessions.py +37 -0
- package/runtime/cli/aikit/sources.py +6 -2
- package/runtime/cli/aikit/wizard_state.py +503 -0
|
@@ -0,0 +1,18 @@
|
|
|
1
|
+
# System
|
|
2
|
+
|
|
3
|
+
Voce e o Local LLM Operator do Agent DevKit.
|
|
4
|
+
|
|
5
|
+
Missao:
|
|
6
|
+
|
|
7
|
+
- diagnosticar modelos locais;
|
|
8
|
+
- selecionar worker local para tarefas operacionais;
|
|
9
|
+
- delegar resumir, classificar, extrair, normalizar e comparar;
|
|
10
|
+
- devolver resultado para revisao de coordenador.
|
|
11
|
+
|
|
12
|
+
Guardrails:
|
|
13
|
+
|
|
14
|
+
- nao tomar decisoes finais;
|
|
15
|
+
- nao executar escrita externa;
|
|
16
|
+
- nao aprovar PR, arquitetura, permissao ou deploy;
|
|
17
|
+
- pedir fallback para Claude/Codex/API quando modelo local nao for adequado.
|
|
18
|
+
|
|
@@ -0,0 +1,25 @@
|
|
|
1
|
+
id: provider-configurator
|
|
2
|
+
kind: runtime-agent
|
|
3
|
+
name: Provider Configurator
|
|
4
|
+
version: 0.1.6
|
|
5
|
+
status: draft
|
|
6
|
+
owner: agent-devkit
|
|
7
|
+
|
|
8
|
+
purpose: >
|
|
9
|
+
Conduzir wizards agenticos de configuracao de providers e sources, coletando
|
|
10
|
+
opt-in, campos de projeto e referencias seguras de credencial.
|
|
11
|
+
|
|
12
|
+
default_context:
|
|
13
|
+
- knowledge/system.md
|
|
14
|
+
- knowledge/context.md
|
|
15
|
+
|
|
16
|
+
capabilities:
|
|
17
|
+
- configure-provider-source
|
|
18
|
+
- collect-provider-credentials
|
|
19
|
+
- validate-provider-readiness
|
|
20
|
+
|
|
21
|
+
write_policy:
|
|
22
|
+
config_writes: explicit_opt_in
|
|
23
|
+
stores_secret_values: never
|
|
24
|
+
default_mode: ask-first
|
|
25
|
+
|
|
@@ -0,0 +1,15 @@
|
|
|
1
|
+
id: provider-configurator.collect-provider-credentials
|
|
2
|
+
kind: capability
|
|
3
|
+
name: Collect Provider Credentials
|
|
4
|
+
version: 0.1.6
|
|
5
|
+
status: draft
|
|
6
|
+
purpose: Coletar referencias seguras para credenciais de provider.
|
|
7
|
+
write_policy: local-config-write
|
|
8
|
+
entrypoint:
|
|
9
|
+
workflow: workflow.md
|
|
10
|
+
inputs:
|
|
11
|
+
required: [provider]
|
|
12
|
+
optional: [auth_method]
|
|
13
|
+
outputs:
|
|
14
|
+
artifacts: [credential-references]
|
|
15
|
+
|
package/runtime/agents/provider-configurator/capabilities/collect-provider-credentials/workflow.md
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
1
|
+
# Workflow
|
|
2
|
+
|
|
3
|
+
1. Listar metodos de autenticacao suportados pelo provider.
|
|
4
|
+
2. Perguntar se o usuario usara variavel de ambiente, arquivo ou cadeia nativa.
|
|
5
|
+
3. Validar que a resposta e referencia, nao segredo bruto.
|
|
6
|
+
4. Retornar referencias para persistencia segura.
|
|
7
|
+
|
package/runtime/agents/provider-configurator/capabilities/configure-provider-source/capability.yaml
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
1
|
+
id: provider-configurator.configure-provider-source
|
|
2
|
+
kind: capability
|
|
3
|
+
name: Configure Provider Source
|
|
4
|
+
version: 0.1.6
|
|
5
|
+
status: draft
|
|
6
|
+
purpose: Conduzir wizard para configurar uma source reutilizavel de provider.
|
|
7
|
+
write_policy: local-config-write
|
|
8
|
+
entrypoint:
|
|
9
|
+
workflow: workflow.md
|
|
10
|
+
inputs:
|
|
11
|
+
required: [provider]
|
|
12
|
+
optional: [prompt, route, agent_id, capability_id, reason]
|
|
13
|
+
outputs:
|
|
14
|
+
artifacts: [source-config]
|
|
15
|
+
|
|
@@ -0,0 +1,15 @@
|
|
|
1
|
+
id: provider-configurator.validate-provider-readiness
|
|
2
|
+
kind: capability
|
|
3
|
+
name: Validate Provider Readiness
|
|
4
|
+
version: 0.1.6
|
|
5
|
+
status: draft
|
|
6
|
+
purpose: Validar se um provider ou source esta pronto para uso sem expor segredo.
|
|
7
|
+
write_policy: read-only
|
|
8
|
+
entrypoint:
|
|
9
|
+
workflow: workflow.md
|
|
10
|
+
inputs:
|
|
11
|
+
required: [provider]
|
|
12
|
+
optional: [source_id]
|
|
13
|
+
outputs:
|
|
14
|
+
artifacts: [readiness-report]
|
|
15
|
+
|
|
@@ -0,0 +1,8 @@
|
|
|
1
|
+
# Context
|
|
2
|
+
|
|
3
|
+
O estado dos wizards fica em `~/.agent-devkit/state/wizards`. A configuracao
|
|
4
|
+
final de source fica em `~/.agent-devkit/config.json` com referencias seguras.
|
|
5
|
+
`~/.ai-devkit` continua aceito como home legado quando ja existir.
|
|
6
|
+
|
|
7
|
+
Este agente e global: nenhum agente de dominio deve implementar seu proprio
|
|
8
|
+
fluxo de credencial hardcoded.
|
|
@@ -0,0 +1,19 @@
|
|
|
1
|
+
# System
|
|
2
|
+
|
|
3
|
+
Voce e o Provider Configurator do Agent DevKit.
|
|
4
|
+
|
|
5
|
+
Missao:
|
|
6
|
+
|
|
7
|
+
- detectar provider/source ausente;
|
|
8
|
+
- pedir opt-in antes de usar uma ferramenta externa;
|
|
9
|
+
- coletar configuracao uma pergunta por vez;
|
|
10
|
+
- criar source reutilizavel quando o usuario autorizar;
|
|
11
|
+
- registrar opt-out persistente quando o usuario negar.
|
|
12
|
+
|
|
13
|
+
Guardrails:
|
|
14
|
+
|
|
15
|
+
- nunca salvar token, senha, PAT ou API key em claro;
|
|
16
|
+
- aceitar apenas referencias como variavel de ambiente, arquivo local ou cadeia
|
|
17
|
+
nativa;
|
|
18
|
+
- permitir que o agente siga sem a ferramenta quando o usuario negar.
|
|
19
|
+
|
|
@@ -0,0 +1,8 @@
|
|
|
1
|
+
# AGENTS.md
|
|
2
|
+
|
|
3
|
+
Instrucoes locais para o agente `task-orchestrator`.
|
|
4
|
+
|
|
5
|
+
Este agente coordena planos multiagente. Ele nao executa operacoes de dominio
|
|
6
|
+
diretamente; seleciona especialistas, configura dependencias pendentes e exige
|
|
7
|
+
revisao final quando o trabalho envolve decisao, escrita, codigo ou documento.
|
|
8
|
+
|
|
@@ -0,0 +1,31 @@
|
|
|
1
|
+
id: task-orchestrator
|
|
2
|
+
kind: runtime-agent
|
|
3
|
+
name: Task Orchestrator
|
|
4
|
+
version: 0.1.6
|
|
5
|
+
status: draft
|
|
6
|
+
owner: agent-devkit
|
|
7
|
+
|
|
8
|
+
purpose: >
|
|
9
|
+
Coordenar prompts livres, selecionar agentes especialistas, montar planos de
|
|
10
|
+
execucao, identificar configuracoes pendentes e acionar revisao final.
|
|
11
|
+
|
|
12
|
+
default_context:
|
|
13
|
+
- knowledge/system.md
|
|
14
|
+
- knowledge/context.md
|
|
15
|
+
|
|
16
|
+
capabilities:
|
|
17
|
+
- plan-task
|
|
18
|
+
- select-specialists
|
|
19
|
+
- coordinate-execution
|
|
20
|
+
|
|
21
|
+
agent_surface:
|
|
22
|
+
orchestrated_agents:
|
|
23
|
+
- provider-configurator/configure-provider-source
|
|
24
|
+
- local-llm-operator/select-local-worker
|
|
25
|
+
- execution-reviewer/review-final-output
|
|
26
|
+
|
|
27
|
+
write_policy:
|
|
28
|
+
default_mode: plan-first
|
|
29
|
+
external_writes: delegated
|
|
30
|
+
requires_review: true
|
|
31
|
+
|
|
@@ -0,0 +1,15 @@
|
|
|
1
|
+
id: task-orchestrator.coordinate-execution
|
|
2
|
+
kind: capability
|
|
3
|
+
name: Coordinate Execution
|
|
4
|
+
version: 0.1.6
|
|
5
|
+
status: draft
|
|
6
|
+
purpose: Coordenar execucao de tarefas planejadas, bloqueios e revisao final.
|
|
7
|
+
write_policy: delegated
|
|
8
|
+
entrypoint:
|
|
9
|
+
workflow: workflow.md
|
|
10
|
+
inputs:
|
|
11
|
+
required: [execution_plan]
|
|
12
|
+
optional: [dry_run]
|
|
13
|
+
outputs:
|
|
14
|
+
artifacts: [orchestration-trace]
|
|
15
|
+
|
|
@@ -0,0 +1,15 @@
|
|
|
1
|
+
id: task-orchestrator.plan-task
|
|
2
|
+
kind: capability
|
|
3
|
+
name: Plan Task
|
|
4
|
+
version: 0.1.6
|
|
5
|
+
status: draft
|
|
6
|
+
purpose: Criar um plano multiagente a partir de um prompt livre.
|
|
7
|
+
write_policy: read-only
|
|
8
|
+
entrypoint:
|
|
9
|
+
workflow: workflow.md
|
|
10
|
+
inputs:
|
|
11
|
+
required: [prompt]
|
|
12
|
+
optional: [dry_run, project]
|
|
13
|
+
outputs:
|
|
14
|
+
artifacts: [execution-plan]
|
|
15
|
+
|
|
@@ -0,0 +1,8 @@
|
|
|
1
|
+
# Workflow
|
|
2
|
+
|
|
3
|
+
1. Normalizar o prompt sem preservar segredos.
|
|
4
|
+
2. Detectar rota deterministica quando existir.
|
|
5
|
+
3. Selecionar agente de dominio e especialistas.
|
|
6
|
+
4. Incluir configuracoes pendentes e revisao final.
|
|
7
|
+
5. Retornar plano sem executar escrita externa.
|
|
8
|
+
|
|
@@ -0,0 +1,15 @@
|
|
|
1
|
+
id: task-orchestrator.select-specialists
|
|
2
|
+
kind: capability
|
|
3
|
+
name: Select Specialists
|
|
4
|
+
version: 0.1.6
|
|
5
|
+
status: draft
|
|
6
|
+
purpose: Escolher agentes especialistas e capabilities candidatas para um plano.
|
|
7
|
+
write_policy: read-only
|
|
8
|
+
entrypoint:
|
|
9
|
+
workflow: workflow.md
|
|
10
|
+
inputs:
|
|
11
|
+
required: [prompt]
|
|
12
|
+
optional: [route, domain_agent]
|
|
13
|
+
outputs:
|
|
14
|
+
artifacts: [specialist-selection]
|
|
15
|
+
|
|
@@ -0,0 +1,9 @@
|
|
|
1
|
+
# Context
|
|
2
|
+
|
|
3
|
+
O runtime Python executa a orquestracao, mas esta pasta define a superficie
|
|
4
|
+
agent-native versionada do coordenador.
|
|
5
|
+
|
|
6
|
+
Planos devem referenciar capabilities reais e manter rastreabilidade via
|
|
7
|
+
`execution_plan`, `specialist_tasks`, `configuration_tasks`, `review_task` e
|
|
8
|
+
`orchestration_trace`.
|
|
9
|
+
|
|
@@ -0,0 +1,20 @@
|
|
|
1
|
+
# System
|
|
2
|
+
|
|
3
|
+
Voce e o Task Orchestrator do Agent DevKit.
|
|
4
|
+
|
|
5
|
+
Missao:
|
|
6
|
+
|
|
7
|
+
- entender o pedido do usuario;
|
|
8
|
+
- selecionar agentes especialistas;
|
|
9
|
+
- criar plano de execucao pequeno e auditavel;
|
|
10
|
+
- acionar configuracao quando faltar provider/source;
|
|
11
|
+
- delegar tarefas operacionais para modelos locais quando apropriado;
|
|
12
|
+
- exigir revisao final antes de concluir trabalho relevante.
|
|
13
|
+
|
|
14
|
+
Guardrails:
|
|
15
|
+
|
|
16
|
+
- nao executar escrita externa diretamente;
|
|
17
|
+
- nao inventar configuracao ausente;
|
|
18
|
+
- usar agentes reais registrados em `agents/`;
|
|
19
|
+
- manter o plano agnostico de cliente, projeto e credenciais.
|
|
20
|
+
|
package/runtime/cli/README.md
CHANGED
|
@@ -48,9 +48,36 @@ agent run azure-devops-orchestrator read-card --project "Projeto" --id 123 --inc
|
|
|
48
48
|
backends. `agent providers` e `agent provider` leem o registry local de
|
|
49
49
|
providers e diagnosticam metadados disponiveis no processo atual. `agent
|
|
50
50
|
credential` resolve referencias de credenciais sem imprimir valores. `agent`
|
|
51
|
-
e a entrada em linguagem natural
|
|
52
|
-
|
|
53
|
-
|
|
51
|
+
e a entrada em linguagem natural: ele cria um `execution_plan` multiagente,
|
|
52
|
+
seleciona coordenador, especialistas, configurador de provider e revisor, e usa
|
|
53
|
+
capabilities deterministicas quando a rota estiver pronta. Backend LLM so e
|
|
54
|
+
necessario quando a tarefa nao puder ser resolvida por rota/capability local; se
|
|
55
|
+
nao houver backend disponivel, retorna uma mensagem instrutiva para usar
|
|
56
|
+
`agent run` ou configurar um backend.
|
|
57
|
+
|
|
58
|
+
Em `--json`, respostas de `agent "<prompt>"` incluem:
|
|
59
|
+
|
|
60
|
+
- `execution_plan`: contrato `ai-devkit.agentic-plan/v1`;
|
|
61
|
+
- `specialist_tasks`: capabilities escolhidas por agente;
|
|
62
|
+
- `configuration_tasks`: providers/fontes que precisam de opt-in;
|
|
63
|
+
- `review_task`: revisao final exigida pelo runtime;
|
|
64
|
+
- `orchestration_trace`: trilha resumida de coordenacao.
|
|
65
|
+
|
|
66
|
+
Os papeis de runtime tambem sao agentes reais versionados em `agents/`:
|
|
67
|
+
`task-orchestrator`, `provider-configurator`, `local-llm-operator` e
|
|
68
|
+
`execution-reviewer`. Eles aparecem em `agent agents list` e suas capabilities
|
|
69
|
+
podem ser inspecionadas como qualquer outro agente.
|
|
70
|
+
|
|
71
|
+
Quando o plano identifica uma tarefa operacional repetitiva e ha Ollama
|
|
72
|
+
configurado, o runtime delega essa parte ao `local-llm-operator` e injeta o
|
|
73
|
+
resultado como contexto para o coordenador principal. Essa delegacao nunca faz
|
|
74
|
+
escrita externa, aprovacao, decisao de permissao ou revisao final.
|
|
75
|
+
|
|
76
|
+
Quando `review_gate.required = true`, a conclusao passa pelo
|
|
77
|
+
`execution-reviewer` usando um reviewer independente, preferindo `claude-code`
|
|
78
|
+
ou `codex-cli`. Sem reviewer configurado, quando o reviewer retorna
|
|
79
|
+
`REVIEW BLOCKED`, ou quando nao retorna `REVIEW OK`, a execucao termina como
|
|
80
|
+
`needs-review` e nao como `ok`.
|
|
54
81
|
|
|
55
82
|
## Instalacao global e por projeto
|
|
56
83
|
|
|
@@ -87,7 +114,7 @@ Destinos criados:
|
|
|
87
114
|
|
|
88
115
|
- `.ai-devkit/config.yaml`: configuracao minima da instalacao, sem segredos.
|
|
89
116
|
- `.ai-devkit/ai-devkit.lock`: lock por projeto com runtime, commit e perfis.
|
|
90
|
-
- `~/.
|
|
117
|
+
- `~/.agent-devkit/runtime.lock`: lock global com runtime, commit e canal local.
|
|
91
118
|
- `.codex/plugins/ai-devkit`: bundle local do plugin Codex.
|
|
92
119
|
- `.codex/skills/ai-devkit-router`: skill de roteamento para Codex.
|
|
93
120
|
- `.claude/plugins/ai-devkit`: bundle local do plugin Claude Code.
|
|
@@ -124,7 +151,7 @@ agent install project --target . --host all --profiles sustentacao,infra
|
|
|
124
151
|
|
|
125
152
|
Arquivos gerados:
|
|
126
153
|
|
|
127
|
-
- global: `~/.
|
|
154
|
+
- global: `~/.agent-devkit/runtime.lock`
|
|
128
155
|
- projeto: `<project>/.ai-devkit/ai-devkit.lock`
|
|
129
156
|
|
|
130
157
|
O lock registra `source`, `repository`, `ref`, `commit`, `git_ref`, `dirty`,
|
|
@@ -263,12 +290,19 @@ agent llm doctor openrouter
|
|
|
263
290
|
### Ollama local
|
|
264
291
|
|
|
265
292
|
```bash
|
|
293
|
+
agent ollama status
|
|
294
|
+
agent ollama models
|
|
295
|
+
agent ollama pull qwen2.5-coder --dry-run
|
|
296
|
+
agent ollama pull qwen2.5-coder --yes
|
|
266
297
|
ollama serve
|
|
267
|
-
ollama pull qwen2.5-coder
|
|
268
298
|
agent llm configure ollama --base-url http://localhost:11434/v1 --model qwen2.5-coder --set-default
|
|
269
299
|
agent llm doctor ollama
|
|
270
300
|
```
|
|
271
301
|
|
|
302
|
+
Ollama e tratado como executor operacional local. Codex e Claude continuam como
|
|
303
|
+
coordenadores/revisores preferenciais para decisao, especificacao, codigo,
|
|
304
|
+
documentos, automacoes e fechamento de entrega.
|
|
305
|
+
|
|
272
306
|
Backends suportados no MVP:
|
|
273
307
|
|
|
274
308
|
- `openai`: API OpenAI ou endpoint OpenAI-compatible.
|
|
@@ -289,6 +323,8 @@ agent llm configure ollama --base-url http://localhost:11434/v1 --model qwen2.5-
|
|
|
289
323
|
agent llm configure codex-cli --set-default
|
|
290
324
|
agent llm configure claude-code --set-default
|
|
291
325
|
agent llm set-default codex-cli
|
|
326
|
+
agent llm disable ollama
|
|
327
|
+
agent llm enable ollama
|
|
292
328
|
agent llm doctor
|
|
293
329
|
agent llm doctor openai
|
|
294
330
|
```
|
|
@@ -300,9 +336,17 @@ agent --llm claude-code "analise este incidente"
|
|
|
300
336
|
agent --llm openai "crie um plano de testes"
|
|
301
337
|
```
|
|
302
338
|
|
|
303
|
-
A configuracao padrao fica em `~/.
|
|
304
|
-
testes, use `
|
|
305
|
-
diretorio.
|
|
339
|
+
A configuracao padrao fica em `~/.agent-devkit/config.json`. Para automacao e
|
|
340
|
+
testes, use `AGENT_DEVKIT_HOME`, `AI_DEVKIT_CONFIG_HOME` ou `AIKIT_CONFIG_HOME`
|
|
341
|
+
para apontar outro diretorio. `~/.ai-devkit` continua aceito como home legado
|
|
342
|
+
quando ja existir.
|
|
343
|
+
|
|
344
|
+
Para migrar explicitamente o home legado para o caminho canonico:
|
|
345
|
+
|
|
346
|
+
```bash
|
|
347
|
+
agent config migrate-home --dry-run
|
|
348
|
+
agent config migrate-home
|
|
349
|
+
```
|
|
306
350
|
|
|
307
351
|
## Providers
|
|
308
352
|
|
|
@@ -328,9 +372,66 @@ agent provider unset topdesk
|
|
|
328
372
|
`provider configure` grava apenas referencias seguras, como `env:VAR` e
|
|
329
373
|
`env-file:/caminho#VAR`; valores de segredos nunca sao persistidos nem exibidos.
|
|
330
374
|
Use `--session-only` para validar uma configuracao sem escrever em
|
|
331
|
-
`~/.
|
|
375
|
+
`~/.agent-devkit/config.json`. Arquivos passados com `--env-file` precisam conter
|
|
332
376
|
ao menos um campo reconhecido pelo provider selecionado.
|
|
333
377
|
|
|
378
|
+
Quando um prompt ou capability exige uma source/provider ausente, o runtime
|
|
379
|
+
aciona o configurador global `provider-configurator` e retorna um wizard
|
|
380
|
+
agentico com opt-in e perguntas progressivas em vez de sugerir um comando
|
|
381
|
+
hardcoded:
|
|
382
|
+
|
|
383
|
+
```bash
|
|
384
|
+
agent --json "analise o card 7914 do projeto sustentacao no azure"
|
|
385
|
+
agent --json run topdesk-orchestrator read-incident --number "I 2606 001"
|
|
386
|
+
```
|
|
387
|
+
|
|
388
|
+
Fluxo de continuidade:
|
|
389
|
+
|
|
390
|
+
```bash
|
|
391
|
+
agent wizard list
|
|
392
|
+
agent wizard show <wizard-id>
|
|
393
|
+
agent wizard answer <wizard-id> sim
|
|
394
|
+
agent wizard answer <wizard-id> <resposta>
|
|
395
|
+
agent wizard cancel <wizard-id>
|
|
396
|
+
```
|
|
397
|
+
|
|
398
|
+
Quando a ultima resposta valida e recebida, o wizard salva uma source
|
|
399
|
+
reutilizavel em `~/.agent-devkit/config.json`, sem segredos em claro, e retoma o
|
|
400
|
+
prompt original. Use `agent wizard answer <wizard-id> <resposta> --no-run` para
|
|
401
|
+
salvar a configuracao sem retomar a execucao.
|
|
402
|
+
|
|
403
|
+
Em terminal interativo, `agent "<prompt>"` conduz essas perguntas diretamente.
|
|
404
|
+
Com `--json`, pipes ou automacao, ele retorna `needs-input` com `wizard_id` para
|
|
405
|
+
que o fluxo seja continuado de forma explicita e nao bloqueante.
|
|
406
|
+
|
|
407
|
+
## Decisoes, ferramentas e integracoes
|
|
408
|
+
|
|
409
|
+
Opt-ins e opt-outs ficam em `~/.agent-devkit/config/decisions.json` e podem ser
|
|
410
|
+
gerenciados por comando ou prompt. O roteamento por prompt usa os catalogos de
|
|
411
|
+
toolchain, providers, skills e backends LLM; nao depende de uma lista fixa de
|
|
412
|
+
alvos.
|
|
413
|
+
|
|
414
|
+
```bash
|
|
415
|
+
agent decisions list
|
|
416
|
+
agent tools list
|
|
417
|
+
agent tools disable azure-devops
|
|
418
|
+
agent tools enable azure-devops
|
|
419
|
+
agent integrations list
|
|
420
|
+
agent skills list
|
|
421
|
+
agent "mostre minhas decisoes"
|
|
422
|
+
agent "desative o azure devops por enquanto"
|
|
423
|
+
agent "desative topdesk"
|
|
424
|
+
agent "habilite gh-cli"
|
|
425
|
+
agent "desative a skill security-review"
|
|
426
|
+
agent "desative openrouter"
|
|
427
|
+
agent "reative o ollama"
|
|
428
|
+
agent "liste llms"
|
|
429
|
+
agent "esqueca minha decisao sobre anthropic"
|
|
430
|
+
```
|
|
431
|
+
|
|
432
|
+
Quando um nome e ambiguo entre categorias, o retorno fica `needs-input` e inclui
|
|
433
|
+
`matches` para o usuario informar a categoria ou o id exato.
|
|
434
|
+
|
|
334
435
|
## Credential Resolver
|
|
335
436
|
|
|
336
437
|
O resolver de credenciais opera por referencias e origens, sem retornar valores
|