sovereign-agent

v1 removal manifest (from v0.7)

Maps every v0.7 top-level subsystem to retain as a 1.x concept, rewrite into a smaller module, or remove from Sovereign Agent 1.x. This is not a compatibility table. 1.0 may drop every v0.7 public symbol.

Public names at v0.7 are listed in public-api-v0.7.txt (165 lines including __version__).

Top-level packages and modules

v0.7 path 1.x action Reason
session/ Rewrite Keep inspectable run artifacts around outcome/run terminology; SQLite is the operational ledger
tickets/ Rewrite Smaller evidence/receipt/manifest models
approvals/ Rewrite Digest-bound approvals remain central
providers/ Rewrite Smaller IntelligenceProvider contract; keep Claude/Codex probe knowledge; add Cursor
relay/ Rewrite Actor mailbox with claims, leases, dead letters
registry/ / registries/ Rewrite Actor registry and role bindings, not seat-fleet supervision
repository/ Rewrite (subset) Local git worktrees and ownership checks only
_internal/atomic.py and file writes Rewrite stdlib atomic projections and artifacts
cli/ Replace argparse; no Typer
config.py Replace committed sovereign.toml; no secrets in governance
errors.py Replace typed refusals that teach the invariant
contracts/ Remove as ZeoCore-shaped package 1.x owns its Pydantic models; optional later JSON fixtures without importing ZeoCore
capabilities/ Remove No second capability framework; no ZeoCore runtime dependency
executor/ / planner/ / halves/ / handoff/ Remove v0.2 substrate is not the 1.x teaching core
ipc/ Remove Not needed for local educational outcomes
tools/ / discovery.py Remove @register_tool compatibility window ends with 0.x
orchestrator/ / workers/ / fleet/ Remove No Docker, Podman, SSH, or fleet scheduling in 1.x
api/ Remove No HTTP or Unix-socket control plane
channels/ Remove No Slack, email, webhooks in core
connectors/ Remove from core Store uses a command-connector boundary in reference_organizations/store/
voice/ / memory/ / observability/ Remove Unimplemented or non-teaching surface
plugins/ Remove No plugin marketplace
admission/ Remove as standalone product Authority lives in 1.x policy.py / governance
secrets/ Remove No general-purpose secrets product
artifacts/ Rewrite (subset) SHA-256 manifests for promoted evidence
execution/ Rewrite Bounded subprocess argv arrays; fail-closed receipts
scheduler/ Replace Every, DailyAt, OnEvent only; no croniter
service/ Replace Explicit per-user service install\|status\|uninstall; runtime loop is supervisor
runtime/ / operations/ Remove or absorb 1.x uses .sovereign/ + SQLite, not v0.4 runtime-root fencing as the teaching model
_internal/llm_client.py Remove No OpenAI-compatible SDK client
_internal/isolation.py / landlock Remove from 1.0 core Governance is not sandboxing; record requested provider sandbox, do not claim to strengthen it

v0.7 public concepts (grouped)

Concept 1.x action
Session directory as inspectable state Retain as file artifacts around outcome/run; not the operational database
Atomic file writes and JSONL traces Retain for projections, evidence, and export
Tickets, receipts, manifests, approvals Retain as smaller models
Claude and Codex CLI adapters Rewrite into the smaller provider contract
Cursor Add as first-class adapter (absent in v0.7)
Seat registry and relay Retain as actor registry and mailbox
Git repository/worktree execution Retain local subset
ZeoCore runtime dependency Remove
Direct OpenAI-compatible LLM client Remove
Typer Replace with argparse
croniter / python-dateutil Remove
Fleet, Docker, Podman, SSH Remove from 1.x core
API server and channels Remove
Voice, memory, observability skeletons Remove
161-name root API Replace with ≤30 deliberate names

Pin path

pip install "sovereign-agent<1" remains the documented old-line path.