fix(human_gate): nest prompt_for collected values under additional_input

[ldavidgomez]

Bug

When a human_gate option uses prompt_for to collect free-text input, the collected value is stored in GateResult.additional_input, but the workflow context output previously flattened that value into the gate output root.

This meant downstream templates could not access the value through the intended nested form:

{{ gate.output.additional_input.field }}

Previously, the value was exposed through the flattened path:

{{ gate.output.field }}

That flat shape is unsafe because it can collide with reserved gate output keys. For example, prompt_for: selected could overwrite the selected option value.

Root cause

engine/workflow.py stored the gate result using a dict spread:

self.context.store(
    agent.name,
    {
        "selected": gate_result.selected_option.value,
        **gate_result.additional_input,
    },
)

This placed collected values at gate.output.<field> rather than gate.output.additional_input.<field>.

It also introduced a silent data-corruption risk: if prompt_for used the field name selected, the spread could overwrite the chosen option value with the user's raw text.

The gate_resolved event and _wait_for_web_gate() already use additional_input as a nested key, so the flat spread in workflow context storage was inconsistent with the rest of the implementation.

Fix

Replace the flat spread with an explicit nested key:

self.context.store(
    agent.name,
    {
        "selected": gate_result.selected_option.value,
        "additional_input": gate_result.additional_input,
    },
)

Gates without prompt_for continue to produce:

{"additional_input": {}}

Compatibility note

This changes the previously observable flat output shape for prompt_for values, for example:

{{ gate.output.answer }}

That flat shape appears to be undocumented and unsafe because it can collide with reserved output keys such as selected.

This may affect workflows that were relying on the undocumented flattened form gate.output.<prompt_for_field>.

The canonical downstream reference after this fix is:

{{ gate.output.additional_input.answer }}

Tests added

Test	What it covers
test_human_gate_stores_additional_input_nested_in_context	output.additional_input.answer is populated after prompt_for input
test_human_gate_no_prompt_for_has_empty_additional_input	Gates without prompt_for still produce additional_input: {}
test_human_gate_prompt_for_named_selected_does_not_corrupt_selected	prompt_for: selected does not overwrite the chosen option value

Tests run

uv run pytest tests/test_engine/test_workflow.py -k "human_gate" -v
7 passed

uv run pytest tests/test_gates/ -v
48 passed

Total relevant validation:

55 passed, 0 failed

tests/test_gates/test_dialog.py       18 passed
tests/test_gates/test_human.py        30 passed
tests/test_engine/test_workflow.py     7 passed

Smoke result

CLI smoke confirmed:

Gate output dump: {'output': {'selected': 'provide_answer', 'additional_input': {'answer': 'README.md'}}}

Selected: provide_answer
Answer: README.md

Confirmed template access:

{{ ask_human.output.additional_input.answer }}

renders:

README.md

Known limitation

In context: explicit, the current input projection logic does not support nested shorthand declarations such as:

input:
  - ask_human.additional_input.answer

The working explicit-mode declaration is to include the parent key:

input:
  - ask_human.selected
  - ask_human.additional_input

and then let the template traverse into it:

{{ ask_human.output.additional_input.answer }}

This is a generic context input-projection limitation, not specific to human_gate or prompt_for, and is intentionally left out of this PR to keep the change focused.

[ldavidgomez]

@microsoft-github-policy-service agree

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (main@085b7a5). Learn more about missing BASE report.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #237   +/-   ##
=======================================
  Coverage        ?   86.84%           
=======================================
  Files           ?       64           
  Lines           ?    10798           
  Branches        ?        0           
=======================================
  Hits            ?     9377           
  Misses          ?     1421           
  Partials        ?        0           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.