[dotnet-port-api] Add agent-as-tool argument name/description customization

[PratikDhanave (PratikDhanave)]

Partial implementation of #1139.

agenttool hardcoded a single query string argument with a fixed description. Add ArgName and ArgDescription config options so a caller can rename/re-describe the argument the model sees — mirroring the .NET/Python as-tool argument options.

Change

• Config.ArgName (default "query") and Config.ArgDescription (default "input query to invoke the agent").
• Schema() emits the configured argument name/description; Call() reads the configured key (decoding into a map rather than a fixed struct field).
• Defaults are unchanged, so existing behavior and tests are unaffected.

Scoped to the argument customization; session propagation (the other half of #1139) is left for the design discussion there, since obtaining the caller session at Call time needs a mechanism the tool interface does not currently expose.

Test

• TestNew_CustomArgNameAndDescription: a custom ArgName:"task" is reflected in the schema (and the default query is gone), and Call with {"task":...} forwards the value to the agent. Fails to build before the change (fields absent), passes after.

[Copilot]

Copilot review overview

🟡 Changes recommended

Invalid non-string argument values are silently converted into blank queries instead of returning decoding errors.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 1

Open (1)

• Preserve string type validation for configured tool arguments · New

What changed in this PR

Adds configurable argument names and descriptions to agenttool, preserving existing defaults.

Changes:

• Added ArgName and ArgDescription configuration.
• Updated schema generation and argument forwarding.
• Added custom-argument tests.

File	Description
tool/​agenttool/​agenttool.go	Implements configurable argument metadata and invocation.
tool/​agenttool/​agenttool_test.go	Tests custom schema and argument forwarding.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[github-actions]

Scope: public API, user-visible behavior

Changed Go contract: agenttool.Config gains two new exported fields, ArgName (default "query") and ArgDescription (default "input query to invoke the agent"), used by Schema() (property name/description, required) and Call() (map-key lookup instead of a fixed Query struct field). Defaults are unchanged from pre-PR behavior.

Upstream evidence reviewed:

• Python python/packages/core/agent_framework/_agents.py, Agent.as_tool() (~line 641): exposes arg_name: str = "task" and arg_description: str | None = None (defaults to f"Task for {tool_name}"), builds input_schema["properties"][arg_name] and required=[arg_name] — the same customization concept the Go PR adds.
• .NET dotnet/src/Microsoft.Agents.AI/AgentExtensions.cs, AIAgentExtensions.AsAIFunction() (~line 66-86): the parameter is a hardcoded local-function parameter named query with a fixed [Description("Input query to invoke the agent.")] attribute; AIFunctionFactoryOptions only lets callers override the function's Name/Description, not the argument name/description. .NET currently has no equivalent argument-customization knob.

Result: aligned (with one non-blocking note). The Go change is a legitimate, additive port of Python's arg_name/arg_description option onto the existing .NET-mirrored default ("query" / "input query to invoke the agent"), and defaults are preserved so existing callers are unaffected. Note for maintainers: the PR description says this "mirrors the .NET/Python as-tool argument options," but .NET does not actually expose an equivalent option today — only Python does. This isn't a parity defect (Go is simply ahead of .NET here, matching Python), just a description nit worth correcting for accuracy. No behavioral or default divergence was found relative to either upstream implementation.

Since this PR adds new exported Config fields, flagging for a public API review via public-api-change. No cross-repo consistency issues found, so also marking parity-approved.

Generated by Go API Consistency Review Agent · copilot · auto · 66.4 AIC · ⌖ 5.97 AIC · ⊞ 9.2K · ◷