lhumina_code/hero_logic

Fork 0

Phase 2: Examples use typed flow inputs #6

New issue

Open

opened 2026-04-19 16:32:20 +00:00 by timur · 2 comments

timur commented

2026-04-19 16:32:20 +00:00

Owner

Context

Phase 1 (#5) introduces typed flow inputs on Workflow. Currently Example.input_data is a free-form JSON string — whatever the user typed into the run form as raw JSON. Once inputs are declared with types, examples should be structured fixtures: one value per declared input.

Goal

Each Example is a named set of values, one per flow input. Loading an example populates the play-start form field-by-field. Saving an example captures the current form values.

Depends on

#5 must be merged first (needs Workflow.inputs: [FlowInput] and typed run form)

OSchema changes

File: crates/hero_logic/schemas/logic/logic.oschema

Replace input_data on Example:

# Example is a named input fixture for a workflow. [rootobject]
Example = {
    sid: str
    workflow_sid: str               # parent workflow
    workflow_version_sid: str       # optional pin to a specific version
    name: str
    description: str
    input_values: {str: str}        # map from flow input name to serialized JSON value
    created_at: int
    updated_at: int
    # REMOVED: input_data (free-form JSON string)
}

Why {str: str} with serialized JSON values: OSchema's map-of-Value is more complex than needed; storing each value as a JSON-encoded string keeps it simple and the Python/JS sides parse per-input by looking at the input's declared type.

Code changes

`crates/hero_logic/src/logic/server/rpc.rs`

example_create(workflow_sid, name, description, input_values) -> Example: validates that provided input_values keys are a subset of the workflow's declared input names; each value can be parsed according to the input's declared type. Unknown keys are dropped with a warning.
example_apply(example_sid) -> {input_values: {str: str}}: fetches the example and returns its values — useful for the UI to load them.
Existing play_start optionally accepts example_sid instead of input_data; if given, resolves the example's input_values and composes them into the input_data JSON.

Migration

For existing Example records with input_data:

Detect legacy shape on read; parse the JSON; best-effort map each top-level key to input_values[key] (values serialized back to JSON strings).
Migrated examples write back on next save.
Provide one-shot example_migrate_legacy(example_sid) RPC.

UI changes

File: crates/hero_logic_ui/templates/workflow_editor.html + JS

"Save as example" modal: instead of a free-form JSON textarea, capture values from the current play-start form fields (one per declared input). Save as input_values: {name: JSON.stringify(value)}.
"Load example" button on each example row: populates play-start form fields from example.input_values (parse per input type).
Example card / list item: shows name + a compact preview of input values (e.g., prompt="ping hero_proc", num_tests=3).
Example validation: if the workflow's input schema changed after the example was created, show a warning icon indicating stale inputs (e.g., input foo no longer exists, or required input bar was added after).

Acceptance criteria

Example record has input_values: {str: str} field; input_data removed
Creating an example from the UI captures structured per-input values
Loading an example populates each play-start form field correctly
Legacy examples (with input_data) still load (auto-migrated on read)
Validation warning shown when an example's keys don't match the workflow's current input schema
Example count per workflow shown in the editor

Out of scope

Example versioning (pinning an example to a specific workflow version — could be done via workflow_version_sid but not strictly required for this phase)
Import/export examples as .json files
Bulk example operations

## Context Phase 1 (#5) introduces typed flow inputs on Workflow. Currently `Example.input_data` is a free-form JSON string — whatever the user typed into the run form as raw JSON. Once inputs are declared with types, examples should be structured fixtures: one value per declared input. ## Goal Each Example is a named set of values, one per flow input. Loading an example populates the play-start form field-by-field. Saving an example captures the current form values. ## Depends on - **#5 must be merged first** (needs `Workflow.inputs: [FlowInput]` and typed run form) ## OSchema changes File: `crates/hero_logic/schemas/logic/logic.oschema` Replace `input_data` on `Example`: ``` # Example is a named input fixture for a workflow. [rootobject] Example = { sid: str workflow_sid: str # parent workflow workflow_version_sid: str # optional pin to a specific version name: str description: str input_values: {str: str} # map from flow input name to serialized JSON value created_at: int updated_at: int # REMOVED: input_data (free-form JSON string) } ``` Why `{str: str}` with serialized JSON values: OSchema's map-of-Value is more complex than needed; storing each value as a JSON-encoded string keeps it simple and the Python/JS sides parse per-input by looking at the input's declared `type`. ## Code changes ### `crates/hero_logic/src/logic/server/rpc.rs` - `example_create(workflow_sid, name, description, input_values) -> Example`: validates that provided `input_values` keys are a subset of the workflow's declared input names; each value can be parsed according to the input's declared `type`. Unknown keys are dropped with a warning. - `example_apply(example_sid) -> {input_values: {str: str}}`: fetches the example and returns its values — useful for the UI to load them. - Existing `play_start` optionally accepts `example_sid` instead of `input_data`; if given, resolves the example's `input_values` and composes them into the input_data JSON. ### Migration For existing Example records with `input_data`: - Detect legacy shape on read; parse the JSON; best-effort map each top-level key to `input_values[key]` (values serialized back to JSON strings). - Migrated examples write back on next save. - Provide one-shot `example_migrate_legacy(example_sid)` RPC. ## UI changes File: `crates/hero_logic_ui/templates/workflow_editor.html` + JS - **"Save as example" modal**: instead of a free-form JSON textarea, capture values from the current play-start form fields (one per declared input). Save as `input_values: {name: JSON.stringify(value)}`. - **"Load example" button** on each example row: populates play-start form fields from `example.input_values` (parse per input type). - **Example card / list item**: shows name + a compact preview of input values (e.g., `prompt="ping hero_proc", num_tests=3`). - **Example validation**: if the workflow's input schema changed after the example was created, show a warning icon indicating stale inputs (e.g., input `foo` no longer exists, or required input `bar` was added after). ## Acceptance criteria - [ ] `Example` record has `input_values: {str: str}` field; `input_data` removed - [ ] Creating an example from the UI captures structured per-input values - [ ] Loading an example populates each play-start form field correctly - [ ] Legacy examples (with `input_data`) still load (auto-migrated on read) - [ ] Validation warning shown when an example's keys don't match the workflow's current input schema - [ ] Example count per workflow shown in the editor ## Out of scope - Example versioning (pinning an example to a specific workflow version — could be done via `workflow_version_sid` but not strictly required for this phase) - Import/export examples as .json files - Bulk example operations

timur referenced this issue

2026-04-19 16:33:13 +00:00

Phase 3: Benchmark flow — statistics per workflow version #7

timur referenced this issue

2026-04-19 16:34:07 +00:00

Phase 4: Request-based version router — pick_version_flow #8

timur referenced this issue from a commit

2026-04-19 17:01:27 +00:00

feat(phase1): workflow versioning + typed flow inputs (#5)

timur referenced this issue from a commit

2026-04-19 17:28:45 +00:00

feat(phase2): Example uses typed input_values (#6)

timur commented

2026-04-19 17:29:41 +00:00

Author

Owner

Backend landed: `671d026`

Phase 2 schema + RPC changes are in. UI changes (per-input form fields in save/load modals, structured preview) remain.

What's live

OSchema:

Example.input_values: JSON-serialized {input_name: value} map
Example.input_data: kept for legacy (auto-migrated on fetch)
Example.workflow_version_sid: optional pin

RPC methods:

example_upsert(sid, workflow_sid, workflow_version_sid, name, description, input_values_json) — validates keys against Workflow.inputs, rejects unknowns with a clear error
example_fetch(sid) — auto-migrates legacy examples (copies input_data → input_values when the latter is empty)
example_to_input_data(example_sid) — new helper that resolves input_values into a play_start-ready input JSON, merging in declared defaults for any unset optional inputs

Verified

example_upsert({prompt:'ping hero_proc'}) → sid ok ✓
example_upsert({prompt:'x', unknown_key:'y'}) → "Unknown input keys: ['unknown_key']" ✓
example_to_input_data(sid) → '{\"prompt\":\"ping hero_proc\"}' ✓
example_fetch(sid) → returns input_values + workflow_version_sid ✓

Remaining for this issue

Schema: input_values, workflow_version_sid
RPC: validation, auto-migration, to_input_data helper
UI: "Save as example" modal captures per-input form values
UI: "Load example" button fills play-start fields per input type
UI: example row shows structured preview (prompt=..., num_tests=...)
UI: stale-schema warning when example keys don't match current inputs

Moving on to Phase 3 (#7) next — benchmark flow.

## Backend landed: 671d026 Phase 2 schema + RPC changes are in. UI changes (per-input form fields in save/load modals, structured preview) remain. ### What's live **OSchema:** - `Example.input_values`: JSON-serialized `{input_name: value}` map - `Example.input_data`: kept for legacy (auto-migrated on fetch) - `Example.workflow_version_sid`: optional pin **RPC methods:** - `example_upsert(sid, workflow_sid, workflow_version_sid, name, description, input_values_json)` — validates keys against `Workflow.inputs`, rejects unknowns with a clear error - `example_fetch(sid)` — auto-migrates legacy examples (copies `input_data` → `input_values` when the latter is empty) - `example_to_input_data(example_sid)` — new helper that resolves `input_values` into a `play_start`-ready input JSON, merging in declared defaults for any unset optional inputs ### Verified ```python example_upsert({prompt:'ping hero_proc'}) → sid ok ✓ example_upsert({prompt:'x', unknown_key:'y'}) → "Unknown input keys: ['unknown_key']" ✓ example_to_input_data(sid) → '{\"prompt\":\"ping hero_proc\"}' ✓ example_fetch(sid) → returns input_values + workflow_version_sid ✓ ``` ### Remaining for this issue - [x] Schema: `input_values`, `workflow_version_sid` - [x] RPC: validation, auto-migration, to_input_data helper - [ ] **UI: "Save as example" modal captures per-input form values** - [ ] **UI: "Load example" button fills play-start fields per input type** - [ ] **UI: example row shows structured preview (`prompt=..., num_tests=...`)** - [ ] **UI: stale-schema warning when example keys don't match current inputs** Moving on to Phase 3 (#7) next — benchmark flow.

timur referenced this issue

2026-04-19 18:08:06 +00:00

Phase 4: Request-based version router — pick_version_flow #8

timur referenced this issue from a commit

2026-04-19 18:25:49 +00:00

feat(phase2-ui): typed example fields + input_values save/load (#6)

timur commented

2026-04-19 18:26:03 +00:00

Author

Owner

Phase 2 UI landed (commit `c062bbf`)

Example save/load now uses structured input_values:

Save-as-example modal renders typed form fields (one per declared FlowInput) when the workflow has declared inputs; falls back to the legacy capture-current-start-node-run_inputs behavior for pre-Phase-2 workflows.
example_upsert now submits input_values_json + workflow_version_sid (full Phase 2 signature) so saved Examples bind to declared inputs and pin to the version that was current at save time.
loadExampleInputs prefers ex.input_values over legacy ex.input_data when applying values into the start-node run_inputs.
Extracted collectTypedInputValues(inputs, prefix) JS helper that both the Start-play and Save-as-example modals share — single source of truth for type coercion (string/number/integer/boolean/object/array) and required-field validation.

Verified end-to-end:

example_upsert("ping","gpt-4o-mini") against workflow 00dz/version 00e0 → saved Example 00e1 with input_values="{...}" and the legacy input_data left empty.
example_to_input_data(00e1) → {"model":"gpt-4o-mini","prompt":"ping"} — ready to feed into play_start.
example_fetch(00e1) returns both input_values and workflow_version_sid so the UI can warn on stale-schema mismatch.

Still open for this issue:

Detect and visibly flag when a loaded Example was saved against a version whose declared inputs have since changed (mismatch warning in the sidebar).
Bulk-apply: one-click "run all examples against current version" (bridges to Phase 3 benchmarking).

### Phase 2 UI landed (commit c062bbf) Example save/load now uses structured `input_values`: - **Save-as-example modal** renders typed form fields (one per declared `FlowInput`) when the workflow has declared inputs; falls back to the legacy capture-current-start-node-run_inputs behavior for pre-Phase-2 workflows. - `example_upsert` now submits `input_values_json` + `workflow_version_sid` (full Phase 2 signature) so saved Examples bind to declared inputs and pin to the version that was current at save time. - `loadExampleInputs` prefers `ex.input_values` over legacy `ex.input_data` when applying values into the start-node `run_inputs`. - Extracted `collectTypedInputValues(inputs, prefix)` JS helper that both the Start-play and Save-as-example modals share — single source of truth for type coercion (string/number/integer/boolean/object/array) and required-field validation. **Verified end-to-end:** - `example_upsert("ping","gpt-4o-mini")` against workflow `00dz`/version `00e0` → saved Example `00e1` with `input_values="{...}"` and the legacy `input_data` left empty. - `example_to_input_data(00e1)` → `{"model":"gpt-4o-mini","prompt":"ping"}` — ready to feed into `play_start`. - `example_fetch(00e1)` returns both `input_values` and `workflow_version_sid` so the UI can warn on stale-schema mismatch. **Still open for this issue:** - Detect and visibly flag when a loaded Example was saved against a version whose declared inputs have since changed (mismatch warning in the sidebar). - Bulk-apply: one-click "run all examples against current version" (bridges to Phase 3 benchmarking).

No labels

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

lhumina_code/hero_logic#6

No description provided.

Rows
Columns