v0.4 — Add Agent Action Boundary before Execution Boundary

[mattgallant001]

Summary

This PR introduces an Agent Action Boundary that interprets and evaluates the legitimacy of requested actions before repository mutations reach the execution boundary.

Previously, the DevWedge evaluated only repository mutations (diff surfaces).
This change adds a pre-execution interpretive stage that evaluates:

• actor identity

• signal surface trust tier

• requested action type

• required authority

before mutation classification occurs.

This establishes a two-stage interpretation model:

Signal
↓
Agent Action Boundary
↓
Action Decision Artifact
↓
Execution Boundary
↓
Mutation Classification
↓
Meaning Artifact

---

Key Capabilities Added

Action Classification

Introduces deterministic classification of requested actions such as:

• package.publish

• release.create

• workflow.modify

• dependency.modify

---

Signal Surface Trust Evaluation

Signal surfaces are evaluated using a registry:

Examples:

Surface	Trust Tier
github.issue.title	untrusted
github.issue.comment	untrusted
github.release.metadata	constrained
github.repo_file.protected	trusted

---

Actor Profile Resolution

Actor identities are resolved using a catalog that defines:

• actor type

• identity confidence

• maximum authority

• allowed intents

• forbidden actions

---

Action Legitimacy Decision

Actions are evaluated against:

• actor authority envelope

• signal trust tier

• declared intent

• required authority

Possible outcomes:

allow
deny
escalate
advisory

---

Action Decision Artifact

The action boundary emits a structured artifact describing the interpretation result:

action_decision_artifact.json

This artifact records:

• actor identity

• signal surface

• requested action

• decision outcome

• evaluation reasons

---

Example Interpretations

Denied

Action: package.publish
Signal: github.issue.title
Actor: automation
Decision: deny
Reason: ACTOR_FORBIDDEN_ACTION

---

Allowed

Action: release.create
Signal: github.release.metadata
Actor: maintainer
Decision: allow

Execution proceeds to the mutation boundary.

---

Canon Updates

This PR also introduces canonical schemas for the new interpretation artifacts:

sp.action_decision_artifact.v1
sp.actor_profile.v1
sp.agent_action_request.v1
sp.signal_surface_catalog.v1

These schemas define the governance structure for action boundary interpretation.

---

Architectural Impact

This PR expands DevWedge from a single mutation boundary to a two-stage interpretive boundary:

Agent Action Boundary
↓
Execution Boundary

This enables deterministic interpretation of both:

• actions proposed by automation

• repository mutations produced by those actions

---

Status

DevWedge prototype now demonstrates:

• agent action boundary

• execution boundary

• signal surface trust tiers

• actor authority envelopes

• deterministic legitimacy evaluation

• structured interpretation artifacts

---

Why This Matters

Automation and AI agents increasingly propose actions:

• publishing packages

• triggering releases

• modifying CI workflows

We typically inspect what changed.

We rarely interpret whether the action itself was legitimate.

This change introduces an interpretive boundary between signals and execution, enabling deterministic legitimacy evaluation before automation actions proceed.

---