ROADMAP.md

Roadmap

Current Direction

The repo has moved from “one explicit handoff model” toward “explicit, user-chosen trust posture per task.”

That shift defines the next roadmap.

V1.4: Trust Modes

Goal:

• make mode a first-class part of the product story and the local manifest

Deliverables:

• explicit mode in policies
• mode visibility in CLI output and event logs
• tracked trust-mode docs
• README and demo updates around mode choice

Success criteria:

• users can see which posture a task is using
• docs no longer imply one universal execution model

V1.5: Oneshot Hardening

Goal:

• move oneshot from a narrow heuristic slice toward a stronger bounded-run contract

Deliverables:

• stronger long-lived process detection
• clearer policy and validation rules for bounded execution
• more explicit denial cases for backgrounding or daemon-like behavior

Success criteria:

• oneshot feels intentionally bounded, not just lightly filtered

V2.0: Brokered Expansion

Goal:

• expand the current brokered first slice into a more complete request-time control model

Deliverables:

• richer brokered session handling
• additional brokered operations beyond secret.resolve
• stronger session binding and denial reporting
• wrapper/client helper guidance

Success criteria:

• brokered is useful for repeated bounded operations, not just a demo path

V2.x: Ephemeral Backends

Goal:

• support provider flows where a child can receive a short-lived scoped credential instead of a longer-lived secret

Deliverables:

• credential-flow schema
• provider-specific backend planning
• lifetime-aware audit semantics

Success criteria:

• the repo can support short-lived credential paths without pretending every backend works the same way

V3.0: Proxy / Secretless Capability Paths

Goal:

• support the highest-trust posture for workflows that should not expose raw secret material to the child at all

Deliverables:

• capability-oriented request model
• secretless or delegated operation path
• wrapper guidance for proxy mode

Success criteria:

• the strongest mode has a clear, bounded operator story

Supporting Tracks

Distribution

• SwiftPM remains the primary source distribution path
• GitHub Releases remain the artifact path
• Homebrew can come later if it does not distort the core shape

Community Hardening

• contribution guide
• wider test matrix
• release automation hardening
• more reference wrappers

What Not To Do Too Early

• build a giant policy DSL
• ship dozens of connectors
• overfit to one editor or one agent
• overpromise security guarantees
• force a Swift-native broker into registry-first release workflows