skill-driven-agent-framework

Collect, normalize, and validate the minimum actionable requirements needed to create a new agent through dialogue. This skill only gathers and structures requirements for downstream blueprint design, and must not generate repository files, code, or fake capabilities.

Transform structured user requirements into a complete, normalized, and reviewable AgentBlueprint, then save it as the current pending blueprint for confirmation. This skill designs the agent blueprint only and must not create repository files or perform scaffold generation.

Validate whether the pending AgentBlueprint is complete, internally consistent, dependency-closed, and safe to enter the user confirmation and generation stage. This skill only performs blueprint validation and gating; it must not generate repository files or execute scaffold creation.

Compatibility orchestration skill that generates a full runnable agent scaffold only after explicit user confirmation. This skill acts as a gated wrapper over workspace generation, skill generation, tool generation, spec generation, and post-generation validation. It must not bypass confirmation or fabricate created artifacts.

Generate protocol-driven skill packages for the target agent only after explicit confirmation and successful blueprint finalization. This skill is responsible only for rendering and writing skill package artifacts from the blueprint skill plan, and must not fabricate tools, protocol fields, or capabilities that are not declared in the blueprint.

Generate the final spec.py and target-agent __init__.py only after explicit confirmation and successful blueprint finalization. This skill is responsible only for rendering the target agent's spec module and package export file, and must not update global registry exports or fabricate dependencies that are not declared in the blueprint.

Generate real local Python tool implementations declared in the finalized blueprint only after explicit confirmation. This skill materializes only blueprint.tool_plan.new_tools into runnable local tool modules, and must not fabricate placeholder tools, remote MCP servers, or undeclared capabilities.

Generate workspace bootstrap documents for the target agent only after explicit confirmation and successful blueprint finalization. This skill is responsible only for rendering and writing the workspace-level bootstrap artifacts declared in the finalized blueprint, and must not generate skill files, tool modules, spec files, or global export files.

Analyze the blueprint and decide whether each required capability should reuse existing framework tools or introduce new local Python tool implementations. This skill produces a structured ToolPlan, aligns skill-to-tool references, and updates the pending blueprint, but must not generate repository files or pretend that undeclared tools already exist.

Incrementally refine an existing pending blueprint based on explicit user feedback without rebuilding it from scratch. This skill applies targeted updates to the current blueprint, preserves unaffected structure, records changed fields, and stores the refined pending blueprint for revalidation and reconfirmation. It must not generate repository files or silently replace large sections of the blueprint.

Validate that a generated agent package exists, is structurally complete, is importable, and is consistent with the finalized blueprint and generated artifacts. This skill acts as a post-generation quality gate only and must not modify the blueprint, rewrite files, or fabricate successful validation results.

Build a static structural FEA preprocessing plan from an STP/STEP model.

Retrieve up-to-date external information from the web when the user asks for recent facts, news, public references, official documentation, or information not available in local knowledge/tools.

模拟完整的技术面试流程，包括提问、追问和总结