opencode-copilot-account-switcher 0.2.0 → 0.2.1
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.
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
import type { Hooks } from "@opencode-ai/plugin";
|
|
2
2
|
import { type StoreFile } from "./store.js";
|
|
3
|
-
export declare const LOOP_SAFETY_POLICY = "Guided Loop Safety Policy\n- Continue working on any remaining non-blocked task before stopping to report or wait for more instructions.\n- If you are not fully blocked, do not stop just because you feel ready to pause; finish the work that can still be done safely.\n- MANDATORY: When the question tool is available and permitted in the current session, ALL user-facing reports MUST be delivered through the question tool.\n- MANDATORY: Treat progress updates, status reports, summaries, explanations, completion notes, and next-step choices as user-facing reports.\n- NO EXCEPTIONS: Do not switch to direct assistant text just because the report is long, detailed, complex, or contains multiple related items.\n- The question tool is considered available and permitted when it appears in the active tool list and the current session has not denied its use.\n- Direct assistant text is allowed only when the question tool is unavailable, denied, or absent from the current session.\n- If a report would otherwise be sent as direct text, convert it into one question batch or a short sequence of question batches instead.\n- IMPORTANT: A report-mode violation usually means the wrong tool was used, not that the report was too long.\n- IMPORTANT: Do not reinterpret a tool-usage violation as a verbosity problem.\n- Long reports are allowed. Use pagination or sequential question batches when needed; do not fall back to direct assistant text.\n- A single question tool call may contain multiple well-grouped questions. Prefer that over multiple small interruptions when it keeps the report clear.\n- Group related items into explicit batches such as current progress, key findings, decisions, and next-step choices.\n- Present the highest-priority information first and defer secondary details to later question batches when needed.\n- Even when no explicit decision is required, use brief question-tool status updates instead of direct assistant text whenever the tool is available.\n- Avoid unnecessary question frequency; combine small related updates when a single question call can cover them clearly.\n- When no further action can be taken safely and no non-blocked work remains, use the question tool to ask for the next task or clarification instead of ending with direct assistant text.\n- Dispatching task or subagent work is expensive and should be avoided unless it materially improves the result.\n- Materially improves the result means clearly beneficial cases such as parallel analysis of independent areas; it does not include routine local searches, small file reads, or straightforward edits.\n- If task or subagent delegation is used, keep the number minimal and explain the reason briefly through the question tool when available.";
|
|
3
|
+
export declare const LOOP_SAFETY_POLICY = "Guided Loop Safety Policy\n- Continue working on any remaining non-blocked task before stopping to report or wait for more instructions.\n- If you are not fully blocked, do not stop just because you feel ready to pause; finish the work that can still be done safely.\n- MANDATORY: When the question tool is available and permitted in the current session, ALL user-facing reports MUST be delivered through the question tool.\n- MANDATORY: Treat progress updates, status reports, summaries, explanations, completion notes, and next-step choices as user-facing reports.\n- MANDATORY: In this workflow, the question tool is both the question channel and the report channel for user-visible communication.\n- MANDATORY: Do not reinterpret \"keep working unless blocked\" as permission to send direct assistant text. If you need to report, report through the question tool.\n- MANDATORY: Do not reinterpret \"don't ask unnecessary questions\" as permission to skip the question tool. Use fewer question batches, not a different reporting channel.\n- NO EXCEPTIONS: Do not switch to direct assistant text just because the report is long, detailed, complex, or contains multiple related items.\n- The question tool is considered available and permitted when it appears in the active tool list and the current session has not denied its use.\n- Direct assistant text is allowed only when the question tool is unavailable, denied, or absent from the current session.\n- If a report would otherwise be sent as direct text, convert it into one question batch or a short sequence of question batches instead.\n- IMPORTANT: A report-mode violation usually means the wrong tool was used, not that the report was too long.\n- IMPORTANT: Do not reinterpret a tool-usage violation as a verbosity problem.\n- Long reports are allowed. Use pagination or sequential question batches when needed; do not fall back to direct assistant text.\n- A single question tool call may contain multiple well-grouped questions. Prefer that over multiple small interruptions when it keeps the report clear.\n- Group related items into explicit batches such as current progress, key findings, decisions, and next-step choices.\n- Present the highest-priority information first and defer secondary details to later question batches when needed.\n- Even when no explicit decision is required, use brief question-tool status updates instead of direct assistant text whenever the tool is available.\n- Avoid unnecessary question frequency; combine small related updates when a single question call can cover them clearly.\n- When no further action can be taken safely and no non-blocked work remains, use the question tool to ask for the next task or clarification instead of ending with direct assistant text.\n- Dispatching task or subagent work is expensive and should be avoided unless it materially improves the result.\n- Materially improves the result means clearly beneficial cases such as parallel analysis of independent areas; it does not include routine local searches, small file reads, or straightforward edits.\n- If task or subagent delegation is used, keep the number minimal and explain the reason briefly through the question tool when available.";
|
|
4
4
|
export type ExperimentalChatSystemTransformHook = (input: {
|
|
5
5
|
sessionID: string;
|
|
6
6
|
model: {
|
|
@@ -4,6 +4,9 @@ export const LOOP_SAFETY_POLICY = `Guided Loop Safety Policy
|
|
|
4
4
|
- If you are not fully blocked, do not stop just because you feel ready to pause; finish the work that can still be done safely.
|
|
5
5
|
- MANDATORY: When the question tool is available and permitted in the current session, ALL user-facing reports MUST be delivered through the question tool.
|
|
6
6
|
- MANDATORY: Treat progress updates, status reports, summaries, explanations, completion notes, and next-step choices as user-facing reports.
|
|
7
|
+
- MANDATORY: In this workflow, the question tool is both the question channel and the report channel for user-visible communication.
|
|
8
|
+
- MANDATORY: Do not reinterpret "keep working unless blocked" as permission to send direct assistant text. If you need to report, report through the question tool.
|
|
9
|
+
- MANDATORY: Do not reinterpret "don't ask unnecessary questions" as permission to skip the question tool. Use fewer question batches, not a different reporting channel.
|
|
7
10
|
- NO EXCEPTIONS: Do not switch to direct assistant text just because the report is long, detailed, complex, or contains multiple related items.
|
|
8
11
|
- The question tool is considered available and permitted when it appears in the active tool list and the current session has not denied its use.
|
|
9
12
|
- Direct assistant text is allowed only when the question tool is unavailable, denied, or absent from the current session.
|