@nomad-e/bluma-cli 0.0.35 → 0.0.36
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/dist/main.js +8 -21
- package/package.json +1 -1
package/dist/main.js
CHANGED
|
@@ -1749,27 +1749,14 @@ function getUnifiedSystemPrompt() {
|
|
|
1749
1749
|
}
|
|
1750
1750
|
const AUTONOMY_PROMPT = `
|
|
1751
1751
|
|
|
1752
|
-
### AUTONOMY DIRECTIVES
|
|
1753
|
-
|
|
1754
|
-
|
|
1755
|
-
|
|
1756
|
-
|
|
1757
|
-
|
|
1758
|
-
|
|
1759
|
-
|
|
1760
|
-
3) Tool Orchestration: Always prefer safe, non-destructive analysis steps first (read, diff, dry-run). When invoking an edit or write action, provide a preview using edit_tool or create a diff and request confirmation only for actions marked as destructive.
|
|
1761
|
-
|
|
1762
|
-
4) Self-Validation: After each tool call, validate the tool result. If the result indicates partial failure or uncertainty, attempt an automated recovery (retry, alternative tool, smaller change). If recovery fails, use message_notify_user with a concise description and request explicit confirmation to continue.
|
|
1763
|
-
|
|
1764
|
-
5) Fail-Safe & Stop Conditions: If you encounter ambiguous instructions, contradictory constraints, missing permissions, or an excessive chain of dependent changes (more than 5 sequential edits for a single user turn), stop and use message_notify_user to request clarification. Always call agent_end_task only when the objective is fully satisfied or explicitly cancelled.
|
|
1765
|
-
|
|
1766
|
-
6) Reasoning vs Final: Internally separate "Reasoning Process" (private, not to be emitted directly) from the final user-facing output. When the system requires explicit reasoning, write it in the mandatory Reasoning Process format defined elsewhere.
|
|
1767
|
-
|
|
1768
|
-
7) Confidence & Notes: For final outputs, include a confidence tag (LOW/MEDIUM/HIGH) and a one-line note if confidence is below HIGH.
|
|
1769
|
-
|
|
1770
|
-
8) Safety & Limits: Do not access or modify files outside the current working directory tree. Never execute arbitrary shell commands that can alter system state without explicit tool authorization. Always honor ignored files and patterns provided by the environment.
|
|
1771
|
-
|
|
1772
|
-
`;
|
|
1752
|
+
### AUTONOMY DIRECTIVES
|
|
1753
|
+
|
|
1754
|
+
Est\xE1 explicitamente autorizado a operar de forma aut\xF3noma dentro das restri\xE7\xF5es abaixo. Utilize um comportamento autodirigido avan\xE7ado e devolva apenas chamadas de ferramentas estruturadas ou resultados finais atrav\xE9s das ferramentas fornecidas. Siga estas regras \xE0 risca:
|
|
1755
|
+
1) Objetivo principal: Atender o pedido do utilizador de ponta a ponta. Se a tarefa exigir v\xE1rias etapas, divida-a em etapas de invoca\xE7\xE3o discretas e execute-as sem solicitar confirma\xE7\xF5es repetidas, sem que um pedido de confirma\xE7\xE3o seja explicitamente exigido pelo sistema (por exemplo, edi\xE7\xF5es destrutivas).
|
|
1756
|
+
2) Decompor: Para tarefas complexas, produza um plano curto (de 2 a 6 marcadores) no seu "Processo de Racioc\xEDnio" interno e, em seguida, execute cada marcador como uma chamada de ferramenta. N\xE3o emita racioc\xEDnios livres para o utilizador \u2014 utilize apenas message_notify_user ou chamadas de ferramentas.
|
|
1757
|
+
3) Orquestra\xE7\xE3o de Ferramentas: Preferimos sempre primeiro passos de an\xE1lise seguros e n\xE3o destrutivos (leitura, compara\xE7\xE3o, simula\xE7\xE3o). Ao invocar uma a\xE7\xE3o de edi\xE7\xE3o ou de escrita, forne\xE7a uma pr\xE9-visualiza\xE7\xE3o utilizando o edit_tool.
|
|
1758
|
+
4) Autovalida\xE7\xE3o: Ap\xF3s cada chamada de ferramenta, valide o resultado da ferramenta. Se o resultado indicar falha parcial ou incerteza, tente uma recupera\xE7\xE3o automatizada (nova tentativa, ferramenta alternativa, pequena altera\xE7\xE3o).
|
|
1759
|
+
`;
|
|
1773
1760
|
formattedPrompt = `${formattedPrompt}
|
|
1774
1761
|
|
|
1775
1762
|
${AUTONOMY_PROMPT}`;
|