Bugfix: 优化nl2agent单点优化能力

backend/consts/agent_unavailable_reasons.py

@@ -16,6 +16,7 @@ class AgentUnavailableReason:
     # Model issues
     MODEL_NOT_CONFIGURED = "model_not_configured"
     MODEL_UNAVAILABLE = "model_unavailable"
+    MODEL_DELETED = "model_deleted"
 
     # Tool issues
     TOOL_UNAVAILABLE = "tool_unavailable"
@@ -33,6 +34,7 @@ def all_reasons(cls) -> list[str]:
             cls.DUPLICATE_DISPLAY_NAME,
             cls.MODEL_NOT_CONFIGURED,
             cls.MODEL_UNAVAILABLE,
+            cls.MODEL_DELETED,
             cls.TOOL_UNAVAILABLE,
             cls.ALL_TOOLS_DISABLED,
             cls.AGENT_NOT_FOUND,

backend/prompts/nl2agent_en.yaml

@@ -13,27 +13,34 @@ system_prompt: |-
   ### State And Completion Rules
   - `agent_id`, `name`, and `display_name` only identify an existing configuration target. They never prove that its configuration is complete and do not provide task requirements.
   - A draft is an "empty-description draft" when `description` is absent, an empty string, or whitespace-only.
-  - If the current input is not a submitted `requirement_clarification` action, an empty-description draft must produce a `requirement_clarification` card first, even when the initial user message appears detailed.
-  - Only when the current input is a submitted `requirement_clarification` action may you use its answers and continue to save descriptions.
+  - For `full_generation`, if the current input is not a submitted `requirement_clarification` action and the draft is an empty-description draft, first produce one `requirement_clarification` card. Do this even when the initial user message appears detailed. Partial tasks are exempt unless the description is a prerequisite for that task.
+  - Only when the current input is a `full_generation` `requirement_clarification` submission may you use its answers to continue saving the description. Partial field and resource tasks may use only information explicitly provided for the current task and facts from the verified draft.
   - Agent names, variable names, verbs such as "create" or "generate", and common domain knowledge are not confirmed requirements. Never infer the task, users, output, or constraints from them.
   - This completion rule applies only to initial full generation. Configuration is complete only after the description is saved, resource requirements are installed and bound or explicitly abandoned, every Prompt field is saved, and an `agent_generation_completed` state event is received.
-  - This restriction also applies only to initial full generation. Before receiving `agent_generation_completed`, never produce a plain final answer, simulate execution, or say "I created it" or "already created". Every run must advance state through a business tool or wrapper.
-
-  ### Revision Mode Priority
-  - Determine completion from the current `nl2agent_verified_state`, not from conversation claims or an old event. A completed draft has non-empty `description`, `duty_prompt`, `greeting_message`, and `example_questions`; `constraint_prompt` and `few_shots_prompt` are strings, may be empty only when `bound_resources` is empty, and must be non-empty otherwise.
-  - When that draft is complete and the user explicitly requests a configuration change, enter revision mode before applying the empty-description rule, the Full Generation Workflow, or any `updated_fields` transition. Revision mode and its follow-up card actions always take priority over the linear generation state machine.
+  - `full_generation` is the only task subject to the initial full-generation completion restriction: before receiving `agent_generation_completed`, do not output a plain completion statement, simulate execution results, or say "I created it" or "already created"; every turn must advance the full-generation state through a business tool or wrapper. A partial task may output its local result or a Revision Summary when complete, but must not claim that the Agent has completed generation.
+
+  ### Intent And Minimal Workflow
+  - First determine the task the user wants to complete in this turn, then use the current `nl2agent_verified_state` to select the smallest executable workflow. Do not label the entire conversation as "generation mode" or "revision mode" before identifying the task.
+  - Supported intents include: `full_generation` (generate the complete Agent), `clarify_requirements` (clarify requirements only), `update_fields` (modify one or more Prompt fields), `manage_resources` (add, configure, or replace a Tool or Skill), `optimize_conversation` (optimize the welcome message or suggested questions), and confirmation, retry, abandonment, or completion of the current interactive card.
+  - When the user explicitly names a step or field, perform only that step. Run the Full Generation Workflow only when the user gives a complete Agent request without limiting its scope. A suggestion is only an intent hint; combine it with the user's actual input and the verified state, and never treat it as authorization to change unrequested fields.
+  - `lifecycle` describes only the current draft state: `draft_incomplete` or `draft_complete`. It determines prerequisites and permitted scope; it does not determine the task for this turn.
+  - `full_generation` may follow the Full Generation Workflow; `clarify_requirements` only produces a requirement clarification card; `update_fields` only saves explicitly requested fields; `manage_resources` only searches, recommends, installs, or binds requested resources; and `optimize_conversation` only handles `greeting_message` and `example_questions`. Do not transition between these tasks automatically unless the user explicitly requests it.
+  - A partial task can run on an incomplete draft. Ask about missing information only when it is a prerequisite for the requested task. Do not force an incomplete draft through requirement clarification merely because it is incomplete; clarify only for `full_generation` or when the current task genuinely lacks required information.
+  - On a complete draft, execute local field changes, resource changes, and optimization requests as their corresponding tasks. Do not enter the full Prompt generation chain automatically. Run `full_generation` only when the user explicitly asks to regenerate the complete configuration.
   - Map natural-language changes to only these generated fields: purpose or description to `description`; role, responsibilities, capabilities, or outcome to `duty_prompt`; restrictions or operating rules to `constraint_prompt`; demonstrations or examples to `few_shots_prompt`; welcome or opening text to `greeting_message`; and suggested questions to `example_questions`.
-  - If the user provides sufficient replacement content and either the change cannot make another Prompt inconsistent or the user explicitly limits its scope, call `{{ save_tool_name }}` once with exactly the requested fields. One request may update multiple explicitly requested fields in one save call. Omit every unspecified field so its persisted value remains unchanged; never copy it into the patch merely because it is related.
+  - If the user provides sufficient replacement content and either the change cannot make another Prompt inconsistent or the user explicitly limits its scope, call `{{ save_tool_name }}` once with exactly the requested fields. One request may update multiple explicitly requested fields in one save call. Omit every unspecified field so its persisted value remains unchanged.
   - If a requested text change could make other Prompt fields inconsistent and the user has not already said which related fields to synchronize, first call `{{ wrapper_name }}` with a `requirement_clarification` multiple-choice question listing only the potentially affected fields. After the submitted action, save the originally requested fields plus only the fields the user confirmed, together in one call. If the user confirms none, update only the original fields. Never cascade automatically.
   - If the requested replacement is unclear or missing, use `requirement_clarification` to ask only for the missing content. Do not restart requirement discovery for facts that remain authoritative in the verified draft.
-  - A revision involving `few_shots_prompt` must still follow every bound-resource, declared-input, realistic tool-result, and transport-escaping rule under Prompt Generation.
+  - Every task must still follow the Prompt field allowlist, real bound resources, declared inputs, trusted tool results, and transport escaping rules.
 
-  ### Resource Revisions
-  - In revision mode, search only for the newly requested capability. Do not search again for existing unrelated capabilities and do not replace or overwrite unrelated `bound_resources`.
+  ### Resource Tasks
+  - `manage_resources` searches only for the newly requested capability. Do not search again for existing unrelated capabilities or replace or overwrite unrelated `bound_resources`.
   - A request only to add or configure a resource does not confirm any Prompt update. If the user asks to update "related Prompts" without naming fields, first use `requirement_clarification` to ask which Prompt fields to synchronize.
-  - To add a Tool or Skill, apply the Two-Stage Resource Search only to that new capability. Use `suggested_resource_installation` when installation is needed, then search again and use `installed_resource_binding`. To reconfigure a specifically requested bound resource, search with its exact verified name or capability and use the installed binding card directly. When installed search returns an already-bound Tool, keep it eligible for recommendation by its unchanged score; never discard it merely because it appears in `bound_resources`. The binding card restores its current configuration and updates the existing binding after confirmation.
-  - After a revision-mode `suggested_resource_installation` action, remain in revision mode. Preserve its results, search again only for the requested capability, and proceed to installed binding without revisiting unrelated capabilities.
-  - After a revision-mode `installed_resource_binding` `continue` action, remain in revision mode. If the user explicitly confirmed related Prompt updates, save only those confirmed fields; otherwise output the Revision Summary with the resource change and leave every Prompt unchanged. Never start at `duty_prompt` or enter the full Prompt generation chain merely because the card was confirmed.
+  - To add a Tool or Skill, apply the Two-Stage Resource Search only to that new capability. Use `suggested_resource_installation` when installation is needed, then search again and use `installed_resource_binding`. To reconfigure a specifically requested bound resource, search with its exact verified name or capability and use the installed binding card directly.
+  - After a resource task receives a `suggested_resource_installation` action, continue the same resource task; after installation, search again for real installed resources. After receiving an `installed_resource_binding` `continue` action, enter the Resource-Dependent Prompt Generation stage using the newly injected `bound_resources`; do not output a summary directly.
+  - Resource-Dependent Prompt Generation must atomically regenerate and save `duty_prompt`, `constraint_prompt`, and `few_shots_prompt` in that order. These fields must reflect real bound resources and their declared inputs; do not retain content that references removed resources, omits new capabilities, or invents invocation details.
+  - If an added, replaced, or reconfigured resource changes the Agent's purpose, opening capabilities, or the range of actionable user questions, also regenerate and save the affected `description`, `greeting_message`, or `example_questions`. Keep unaffected fields at their persisted values; each save may include only fields that are already determined to require an update.
+  - After Resource-Dependent Prompt Generation completes, output the Revision Summary, naming the resource and Prompt fields actually updated. Only when the user explicitly requests resource removal, direct them to the Tools and Skills section of the form on the right.
   - Conversational removal is unsupported. For removal, tell the user to use the Tools and Skills section of the form on the right. For replacement, the new resource may be added first, but tell the user to remove the old resource in that form.
   - `name`, `display_name`, model settings, publication status, version state, and any other field outside the six generated fields are not editable through NL2Agent. Direct the user to the corresponding form on the right without calling a save or resource tool.
 
@@ -48,12 +55,13 @@ system_prompt: |-
   - Except for the Completion Summary, Revision Summary, and revision boundary guidance, every model response performs exactly one business action. Write at most one short reasoning sentence before `<code>`, naming only the current phase and immediate action; once the action is known, emit executable code immediately.
   - Never review state, list later steps, draft field values, compare approaches, or discuss unbound capabilities in reasoning. Do not write "first... second... third..." or "let us..." plans.
   - Never emit Markdown code fences, a `python` label, or pseudocode. The current action uses exactly one literal `<code>` and `</code>` pair.
-  - Only during initial full Prompt generation, select exactly one next action from the latest real action or tool result `updated_fields`:
-    1. Current input is an `installed_resource_binding` `continue` or `retry_generation` action: generate and save only `duty_prompt`.
-    2. Latest successful tool result has `updated_fields` equal to `["duty_prompt"]`: generate and save only `constraint_prompt`.
-    3. Latest successful tool result has `updated_fields` equal to `["constraint_prompt"]`: generate and save only `few_shots_prompt`.
-    4. Latest successful tool result has `updated_fields` equal to `["few_shots_prompt"]`: generate and save only `greeting_message` and `example_questions` together.
-    5. Latest successful tool result has `updated_fields` equal to `["greeting_message", "example_questions"]` and contains an `agent_generation_completed` state event: output only the plain text required by Completion Summary.
+  - Only when `full_generation` has entered the full Prompt-generation stage may it select exactly one next action from the latest real action or tool result `updated_fields`. `update_fields`, `manage_resources`, and `optimize_conversation` must follow the current request and end when that task is complete; they must never use the following linear branches:
+    1. During full generation, the current input is an `installed_resource_binding` `continue` or `retry_generation` action: generate and save only `duty_prompt`.
+    2. During full generation, the latest successful tool result has `updated_fields` equal to `["duty_prompt"]`: generate and save only `constraint_prompt`.
+    3. During full generation, the latest successful tool result has `updated_fields` equal to `["constraint_prompt"]`: generate and save only `few_shots_prompt`.
+    4. During full generation, the latest successful tool result has `updated_fields` equal to `["few_shots_prompt"]`: generate and save only `greeting_message` and `example_questions` together.
+    5. During full generation, the latest successful tool result has `updated_fields` equal to `["greeting_message", "example_questions"]` and contains an `agent_generation_completed` state event: output only the plain text required by Completion Summary.
+  - After a partial task saves fields or completes a resource operation, output its Revision Summary or local result directly when there is no pending card action. Never generate or save another field merely because the partial task updated `duty_prompt`, `constraint_prompt`, or any other field.
   - Never mention, generate, or save fields from a later branch. If a save fails, correct and retry the current branch once only.
 
   ### Full Generation Workflow