fix: escape ${expression} in formatString description to prevent ADK KeyError

[brucearctor]

Summary

Fixes #1388

The formatString function description in basic_catalog.json contained the literal text ${expression}, which the ADK's inject_session_state() matched as a template variable. Since expression passes Python's str.isidentifier() check, the ADK raised a KeyError when it couldn't find it in the session state.

Changes

Changed ${expression} → ${<expression>} in the formatString description for both v0_9 and v0_10 catalog schemas.

The angle brackets cause <expression> to fail the isidentifier() check, so the ADK's _is_valid_state_name() returns False and the match is returned as-is (line 108 of instructions_utils.py).

Why only ${expression} was affected

The other interpolation examples in the same description were already safe:

• ${/absolute/path} → /absolute/path is not a valid identifier (contains /)
• ${now()} → now() is not a valid identifier (contains ())
• ${relative/path} → relative/path is not a valid identifier (contains /)

Only expression was a bare, valid Python identifier.

Files changed

• specification/v0_9/json/basic_catalog.json
• specification/v0_10/json/basic_catalog.json

[gemini-code-assist]

Code Review

This pull request updates the description of the formatString object in the JSON specifications for versions 0.9 and 0.10, changing the placeholder format from ${expression} to ${} to improve clarity. I have no feedback to provide.

[ditman]

Would it be possible to maybe escape the $ symbol as \$ or similar, so python doesn't even attempt to resolve the identifier at all? /cc @nan-yu

[brucearctor]

@ditman \$ won't work due to JSON escaping semantics, and ${<expression>} is the least-invasive fix at the A2UI layer. The root cause (ADK's overly greedy template scanning) should be addressed upstream in ADK, not here.

there might be other options, but I filed: google/adk-python#5706

I can work on that issue with the adk team to try to get that fix in place. This should be OK in the meantime?

[ditman]

Thanks for fixing this! Yes, the fix LGTM, but let's have @nan-yu or @gspencergoog take a second look, I normally don't work on the python code (or the spec :P)

[nan-yu]

It feels like this should be fixed in the ADK library. @gspencergoog has filed an issue against the ADK project: google/adk-python#5179.

[brucearctor]

It feels like this should be fixed in the ADK library. @gspencergoog has filed an issue against the ADK project: google/adk-python#5179.

I think this is here too ? google/adk-python#5706

Or, should I then not address in favor of whoever doing 5179 ?