FE-848: Prompt context refinement

[lunelson]

Stack Context

This PR refines FE-848 prompt context and fixture protocol behavior at the top of the current stack.

What?

• Documents live graph observer and fixture workbench protocols.
• Updates SPEC/PLAN around fixture capture and prompt context expectations.
• Tightens seed fixture tests and Brunch app coverage for the refined protocol.

[lunelson]

• FE-852: Prompt composition golden coverage and elicitation driver #207
• FE-XXX: Projection coverage orientation and scoping #206
• FE-847: DX introspection tier 2 #205
• FE-848: Prompt context refinement #204 👈 (View in Graphite)
• FE-845: TUI chrome resolutions pass II #203
• FE-847: DX introspection tier 2 #202
• FE-844: Wire affordance legality to capability readiness #201
• FE-844: Elicitation gaps architecture #197
• FE-845: TUI chrome resolutions pass I #196
• FE-846: Web revamp pass II #195
• FE-825: DX introspection tier 1 #189
• FE-830: Align with design params of main branch #187
• FE-824: Role-safe graph mutations and completion of full stack graph tools #186
• FE-825: Plan first-class developer feedback loops over the pi harness #180
• FE-811: Ship-gate rename residue and live mention autocomplete #179
• FE-823: Elicitation backlog substrate and agenda read-back #178
• FE-811: Elicitor cross-cut groundwork and read/project/render cleanup #177
• FE-809: Project-graph review-set proposal and atomic acceptance #175
• FE-807: Structured response capture into selected-spec graph truth #173
• FE-808: Broaden direct graph-tool proof beyond the A14 happy path #171
• FE-806: Agent prompt-resource composition, runtime manifests, and snapshot contexts #169
• FE-795: Live selected-spec graph observer over web RPC #166
• FE-785: M5 agent-graph-integration #163
• FE-776: Graph Layer preps #162
• FE-744: Document Pi command containment evidence #150
• FE-737: Web shell over the same host (M3) #147
• FE-736: JSONL session viability proof (M2) #146
• FE-735: Mode shell and fixture driver (M1) #144
• FE-702: Walking skeleton — brunch binary + TUI over pi (M0) #142
• next

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

[cursor]

PR Summary

Low Risk
DX and documentation hardening around dev seeding; graph writes still go through CommandExecutor with new validation tests—no product auth or runtime behavior changes beyond refusing the old default seed path.

Overview
Dev seeding is now explicit and workspace-scoped. npm run seed no longer loads every fixture into the shell cwd; it requires --workspace <dir> and --seed <set>/<slug>, validates flags (including rejecting unsafe paths and duplicates), writes only the target .brunch/data.db via seedFixture/CommandExecutor, and prints destination plus set/slug → specId. Missing args exit with usage instead of seeding.

Docs and architecture registers codify the seed-then-npm run dev -- --cwd workbench flow, clarify that dev launch never implies seeding, and add D79-L / I48-L (plus glossary updates) so seeds, workbenches, runs, and scratch stay distinct. The live-graph-observer workbench README switches to npm run seed + npm run dev -- --cwd and documents the TUI sidecar path instead of standalone --mode web.

Tests cover the seed CLI contract and a Brunch RPC tracer that seeds one workbench and proves workspace.selectionState comes from that cwd, not a sibling or shell cwd.

^{Reviewed by Cursor Bugbot for commit 3134c43. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 2cb5b4f. Configure here.}

[Copilot]

Pull request overview

This PR refines the dev seed-fixture protocol to make seeding explicit and workspace-scoped (no ambient “seed everything into current cwd”), and updates tests/docs/spec/plan to reflect and validate that behavior.

Changes:

• Replace the prior “seed all sets into cwd” flow with an explicit runSeedFixturesCli that requires --workspace + --seed <set>/<slug> and seeds exactly one fixture.
• Add CLI contract tests for argument validation and workspace isolation; add an app-level RPC test proving --cwd inspects the targeted workspace DB.
• Update SPEC/PLAN and .fixtures/ documentation to clarify the seeds/workbenches/runs/scratch separation and the explicit seeding workflow.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/graph/seed-fixtures.ts	Introduces explicit seed CLI parsing and single-fixture seeding into a named workspace DB.
src/graph/seed-fixtures.test.ts	Adds CLI-focused tests for safe arg parsing, workspace targeting, and single-fixture behavior.
src/app/brunch.test.ts	Adds an integration-style check that RPC state reflects the targeted --cwd workspace (not shell cwd).
memory/SPEC.md	Refines fixture/seed/workbench protocol and records the I48-L validation progress.
memory/PLAN.md	Updates the dev-seed-fixtures frontier to reflect the folded-in FE-848 hardening slice and acceptance leaves.
.fixtures/workbenches/live-graph-observer/README.md	Documents the new explicit “seed then launch with --cwd” workflow and removes unsupported launch modes.
.fixtures/README.md	Adds canonical seed + workbench usage guidance and clarifies non-implicit seeding.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.