MVP Proposal

MVP objective

Prove that Alloy can turn engineering intent into better Foundry runs without rewriting Foundry.

MVP scope

0. Phoenix/LiveView administration system

Treat Alloy as a user-facing administration product from the beginning. The MVP should have a small Phoenix/LiveView surface for projects, engineering intent records, hypothesis review, Formation Brief review, and Foundry run feedback. This does not mean the integration must be deep in the MVP; it means the human-facing intent model should not be hidden in CLI-only files.

1. Engineering intent records

• Create/edit/list records.
• Fields: capability, threat, expectation, strategy, evidence, tradeoff, status, scope, source.
• Store in PostgreSQL or local SQLite/Postgres-compatible model for early development.
• Export markdown/YAML projection.

2. Assistant-led elicitation

Support three interview modes:

• Why is this here?
• What would make you reject a PR?
• What scar are you avoiding?

3. Lightweight codebase archaeology

Start with static repository inspection (see Codebase Archaeology):

• Directory/module structure.
• Test structure.
• Imports/dependencies.
• Existing gates.
• CHARTER.md, README, AGENTS-like files.

Output hypotheses only.

4. Formation brief generation

Given:

• One product objective or plain-language mission.
• Selected engineering intent records.
• Optional Epilogue interaction reference.

Produce:

• Formation brief YAML.
• Prompt pack text files.
• Suggested Foundry payload.

5. Foundry integration by projection

Do not initially require Foundry code changes. Generate a local package Foundry can consume through existing prompt or strategic prompt mechanisms — the projection mode described in Integration Modes.

Example output directory:

.alloy/
  briefs/
    checkout-payment.yaml
  prompts/
    assessment.md
    triage.md
    planning.md
    execution.md
    retry.md
    summary.md
  gates/
    overlay.hone-gates.json
  projections/
    CHARTER.generated.md

6. Trace feedback import

Start with manual trace ingestion:

alloy trace ingest <event-id>

Summarize:

• Which gates passed.
• Whether retry happened.
• Which intent records were in scope.
• Which evidence remains unknown.

MVP success criteria

The MVP is successful if:

1. A developer can create useful engineering intent records in under 30 minutes.
2. Alloy can discover at least three plausible intent hypotheses from a real codebase.
3. A human can accept/edit/reject those hypotheses quickly.
4. Alloy can compile selected intent into a Foundry-usable prompt package.
5. A Foundry run using Alloy context produces more relevant assessment and planning than a generic strategic_prompt.
6. Alloy can ingest a Foundry trace and report intent satisfaction or gaps.

Up: Delivery Plan

Source: Product Brief §23 (MVP Proposal).