Spec gap: `chain` agent behavior when `steps` and `prompt`/`prompt_reference` are combined is unspecified #20

New issue

Open

opened 2026-05-27 05:27:04 +00:00 by hurui200320 · 0 comments

hurui200320 commented

2026-05-27 05:27:04 +00:00

Member

Problem

§4.7 defines the chain agent with three optional fields: steps, prompt, and prompt_reference. The spec forbids combining prompt and prompt_reference together, but says nothing about combining steps with either prompt field.

The invocation logic describes two independent behaviors:

If prompt/prompt_reference is set: Render the prompt with the input message.
Iterate through steps: Append each step label with -> separator.

When both a prompt and steps are present, it is unclear:

Is the rendered prompt used as the starting value that steps are then appended to?
Or are the prompt and the steps independent outputs that are somehow combined?
What is the prefix behavior — is it "ChainAgent processed: <rendered_prompt> -> step1 -> step2" or something else?

Additionally, the edge case of an empty steps list with no prompt is not addressed. Should the agent return "ChainAgent processed: " (empty suffix) or signal an error?

Fix Needed

Add explicit semantics for the following combinations in §4.7:

prompt (or prompt_reference) + steps: define the exact output format
Neither prompt nor steps provided (all optional, all absent): define fallback behavior
steps: [] (empty list): define whether this is valid and what is returned

References

§4.7 — Chain Agents

## Problem §4.7 defines the `chain` agent with three optional fields: `steps`, `prompt`, and `prompt_reference`. The spec forbids combining `prompt` and `prompt_reference` together, but says nothing about combining `steps` with either prompt field. The invocation logic describes two independent behaviors: 1. **If `prompt`/`prompt_reference` is set**: Render the prompt with the input message. 2. **Iterate through `steps`**: Append each step label with ` -> ` separator. When both a prompt and steps are present, it is unclear: - Is the rendered prompt used as the starting value that steps are then appended to? - Or are the prompt and the steps independent outputs that are somehow combined? - What is the prefix behavior — is it `"ChainAgent processed: <rendered_prompt> -> step1 -> step2"` or something else? Additionally, the edge case of an empty `steps` list with no prompt is not addressed. Should the agent return `"ChainAgent processed: "` (empty suffix) or signal an error? ## Fix Needed Add explicit semantics for the following combinations in §4.7: 1. `prompt` (or `prompt_reference`) + `steps`: define the exact output format 2. Neither prompt nor steps provided (all optional, all absent): define fallback behavior 3. `steps: []` (empty list): define whether this is valid and what is returned ## References - §4.7 — Chain Agents

Rows
Columns

Spec gap: chain agent behavior when steps and prompt/prompt_reference are combined is unspecified #20

Problem

Fix Needed

References

Spec gap: `chain` agent behavior when `steps` and `prompt`/`prompt_reference` are combined is unspecified #20