FE-698: Scenario runner + observer scenario composition

[lunelson]

Stack Context

This FE-698 stack establishes reusable prompt/context machinery and a pre-UI scenario substrate for evaluating agent behavior.

What?

• Adds a deterministic no-provider scenario artifact builder for prompt probes.
• Composes observer-capture scenarios from the typed observer context pack and production observer prompt composition.
• Extracts observer prompt composition into a pure module and makes prompt sources explicit (asset vs composed).

Why?

Brunch needs repeatable prompt probe artifacts before adding real provider/harness execution. This PR proves the artifact shape and observer scenario composition while keeping provider/Pi/runtime authority out of the substrate.

[lunelson]

• FE-705: Plan agent capability CLI frontier #127
• FE-698: On more context pack + spec for agent CLI v1 #114
• FE-698: Prompt/context scenario substrate foundation #111
• FE-698: Agent mutation audit; tool surface design #104
• FE-698: Design/spec/plan augmentation #103
• FE-698: Scenario runner + observer scenario composition #102 👈 (View in Graphite)
• FE-698: Prompt registry + observer context-pack foundation #101
• FE-698: Prompt/context substrate planning #98
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[cursor]

PR Summary

Medium Risk
Medium risk because it refactors the production observer system-prompt composition into a new module and introduces new prompt-loading APIs; any subtle prompt-string change could impact observer extraction behavior.

Overview
Introduces a lightweight scenario runner that builds deterministic, no-provider “probe artifacts” for prompt/context scenarios (currently observer-capture), including stable serialization and validation that prompt assets don’t contain unresolved template variables.

Refactors observer prompt composition by extracting buildObserverSystemPrompt into a new pure observer-prompt.ts module, and extends the prompt loader with getPromptAssetFileName so probe artifacts can record the underlying markdown asset name. Documentation/spec registers are updated to mark the scenario runner as completed/covered (scenario-runner.test.ts).

^{Reviewed by Cursor Bugbot for commit 8d9ab97. Bugbot is set up for automated code reviews on this repo. Configure here.}

[augmentcode]

Review completed. 1 suggestion posted.

Comment augment review to trigger a new review at any time.

Latest approval commit c99d056 is not an ancestor of 8d9ab97, indicating rewritten history after approval

[graphite-app]

Merge activity

• May 8, 10:46 AM UTC: Graphite rebased this pull request, because this pull request is set to merge when ready.