fix(compaction): preserve namespaced tool names in LLM summarizer prompt

[yashrajshuklaaa]

after context compaction the LlmEventSummarizer rewrote conversation history using an LLM.
also the default ADK prompt had no instruction to preserve exact tool names so the LLM abbreviated namespaced tool names like 'kagent__default__agent_name_aa' to just 'agent_name_aa'
when the orchestrator agent resumed after compaction it could not dispatch to the sub-agent because the short name does not match the registered tool name

Fix: inject a kagent-specific default prompt_template that explicitly instructs the summarizer LLM to preserve exact tool/agent names including namespace prefixes.users who supply a custom prompt_template are unaffected (their value takes precedence via 'or')

Fixes #1550

[Copilot]

Pull request overview

This PR fixes a compaction-related failure where the LLM summarizer could abbreviate namespaced tool/agent names (e.g., dropping kagent__...__ prefixes), causing the orchestrator to later reference non-existent tools after context compaction.

Changes:

• Introduces a kagent-specific default compaction prompt that instructs the summarizer to preserve exact registered tool/agent names (including namespace prefixes).
• Updates build_adk_context_configs to use the kagent default prompt when no custom prompt_template is provided.
• Adds unit tests to ensure the default prompt is applied and that a user-provided prompt remains respected.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
python/packages/kagent-adk/src/kagent/adk/types.py	Adds _KAGENT_COMPACTION_PROMPT and uses it as the default summarizer prompt when prompt_template is not provided.
python/packages/kagent-adk/tests/unittests/test_context_config.py	Adds tests verifying the default prompt selection and that custom prompts are not overridden.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

Comments suppressed due to low confidence (1)

python/packages/kagent-adk/tests/unittests/test_context_config.py:207

• This assertion relies on LlmEventSummarizer’s private _prompt_template attribute; using a public prompt_template attribute when available (and falling back) will make the test more resilient across google-adk versions.

        assert events_cfg.summarizer._prompt_template == ""

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated no new comments.

[EItanya]

Thanks so much for the fix. What happens if the user does provide a prompt template, do we always want to append the bit about tool name correctness?

[yashrajshuklaaa]

@EItanya great point
two ways to handle this -
Keep it as is , custom prompt overrides everything. User is responsible for preserving tool names in their prompt
or
always append the tool name warning to whatever prompt is used , even custom ones

i prefer 2nd one cuz appending the warning is a small change but saves users from a confusing failure

[EItanya]

@EItanya great point two ways to handle this - Keep it as is , custom prompt overrides everything. User is responsible for preserving tool names in their prompt or always append the tool name warning to whatever prompt is used , even custom ones

i prefer 2nd one cuz appending the warning is a small change but saves users from a confusing failure

sounds good let's do it

[yashrajshuklaaa]

@EItanya done
PTAL