Skip to content

Littlepowers 1.3.0 — Review Lease and parallel workflows

Choose a tag to compare

View all tags
@clsaa clsaa released this 01 Aug 17:37
· 2 commits to main since this release
v1.3.0
c095354

Important

Littlepowers 1.3.0 writes schema-4 ledgers. A 1.2 runtime cannot read them.
Before rolling back, cancel any open Review Gate, pause or finish the workflow,
restore its matching pre-schema4 schema-3 archive, and only then install 1.2.
Do not hot-replace Littlepowers inside an active task.

Littlepowers 1.3.0 is the stable Review Lease release for Codex, Claude Code,
Qoder, and OpenCode. It keeps planning proportional and model-neutral while
making approved outcomes, review boundaries, and parallel worktree status more
durable and deterministic.

Highlights

  • Review Lease: four explicit policies cover blocking review, a fixed
    implementation mandate, timed fallback, and unattended continuation. Gates
    bind the exact artifact path, bytes, Contract sources, workflow, and revision.
  • Less unnecessary waiting: a fixed Lean/Compact implementation request can
    proceed without another approval stop. An explicitly windowed Claude Code
    gate may arm one bounded same-session callback; no retry, output capture,
    permission bypass, model override, or effort override is added.
  • Parallel project iterations: an optional Project Workflow Index summarizes
    up to 16 explicitly registered worktrees. Every worktree retains its own
    ledger and single writer; hooks never scan or load the index.
  • Outcome integrity: Outcome Lock continues to require complete acceptance
    coverage, approved scope deltas, baseline fidelity, and aggregate completion.
    Implementation is one continuous product scope; tasks, checkpoints, small
    commits, and rollback units remain safe engineering boundaries, not staged
    product slices.
  • Lightweight by default: ordinary routing adds no extra model call, agent,
    repository scan, scheduler, daemon, broad test run, telemetry, or model/effort
    selection. Direct work remains direct; Lean remains brainstorm → plan.
  • Cross-host packaging: the same 11 skills and protocol are validated for
    Codex, Claude Code, Qoder, and OpenCode.

Verification

  • Release commit: c095354a01ff7f09cb539261a44fc1efc9f7d88b.
  • Three independent expert reviews: Critical 0, Important 0, work-unit
    compliance pass, approved-outcome fidelity pass, code quality approve.
  • Local aggregate: 198/198 Python tests passed; all 11 skill validators and all
    four host package surfaces passed.
  • Exact-commit CI:
    Ubuntu/Python 3.9, macOS/Python 3.13, Windows/Python 3.13 (including the Hook
    launcher probe), and Claude Code strict validation all passed.
  • Exact-tag CI:
    the same four jobs passed from annotated tag v1.3.0.
  • Independent local-tag and GitHub-remote-tag clones both passed the 11 skill,
    Codex package, Claude strict, Qoder, OpenCode, identity, and empty-ledger
    doctor smoke checks.

The release does not claim authenticated GPT-5.6 Sol/Ultra, Fable 5, Opus
4.8/5, Claude dynamic-workflow orchestration, or a real Windows Claude runner
end-to-end certification. The protocol is model-neutral; the runner evidence is
deterministic fake-host control and argv coverage plus cross-platform CI.

See the README
for exact-tag installation, updates, rollback, troubleshooting, security, and
host-specific limitations. The detailed integrated evidence is in
evals/results/2026-08-01-v1.3.0-release.md.

Assets 2
Loading