Skip to content

Releases: RNT56/NilCore

v1.1.0

Choose a tag to compare

@github-actions github-actions released this 21 Jun 04:07
3fbc297

What's Changed

  • Phase 13: Trust Ledger + "why did it do that" trace explorer by @RNT56 in #55
  • Phase 13 (languages): code intelligence now reads 19 languages by @RNT56 in #56
  • Phase 13: multi-backend strength-routing (the Trust Ledger goes live) by @RNT56 in #57
  • Phase 13: model-driven work-routing + earned backend selection by @RNT56 in #58
  • release: v1.1.0 — Phase 13 (model-driven routing, earned backend selection, 19 languages, trust/trace) by @RNT56 in #59

Full Changelog: v1.0.1...v1.1.0

v1.0.1 — audit remediation

Choose a tag to compare

@RNT56 RNT56 released this 21 Jun 00:14
624a99d

A post-1.0.0 patch from a full adversarial audit of the merged P11 (verifier-backed artifact factory) + P12 (verified swarm) surface (28.6k lines: 8-reviewer review → independent verify → synthesis).

The audit confirmed the architecture and all seven invariants (I1–I7) are sound, and fixed 20 real defects:

  • 🔴 Blocker — code-pack flag injection: a worker Evidence.Value of -run=^$ ran go test -run=^$, selected zero tests, exited 0, and laundered a green verdict past the verifier (I2 erosion). Fixed with a leading-dash reject + a -- positional separator. 1.0.0 shipped with this (the audit ran post-release).
  • 🟠 8 major — finance keyed checks could never pass (single-quoted $VAR), swarm false-RED on repeat runs, pool cap-sharing on inherited tiers, a dead --egress-allow flag, the absent TipSHA fold seam, and 3 masking tests rewritten to discriminate.
  • 🟡 8 minor + 3 polish — termination rails, board redaction, dead-trace removal, non-recording budget probe, and more.

Every defect except the blocker failed safe (false-RED or inert). make verify + go test -race green; go.mod/go.sum unchanged; default binary still 2 core deps, CGO_ENABLED=0.

Full per-defect detail: #53. Release: #54.

v1.0.0

Choose a tag to compare

@github-actions github-actions released this 20 Jun 17:53
eaddb5d

NilCore v1.0.0 — the agent its constitution describes

NilCore is a tiny, robust, sandboxed coding and research agent: the harness is small, the model is the engine, and trust comes from three disciplines only — the verifier, the sandbox, and the append-only audit log. v1.0.0 accumulates Phases 7–12 + the deferred items D1–D4 + the external-review fixes R1–R3 on top of the [v0.1.0] core (Phases 0–6).

What it does

  • One task or a whole project — a native loop, or an agentic supervisor that spawns role-specialized subagents, integrates their parallel worktrees into one verifier-green tree, and re-plans to convergence. Plus delegated Codex / Claude Code behind one frozen contract.
  • Verified swarm mode (nilcore swarm) — fan hundreds of agents into a bounded in-process pool where every unit produces a typed artifact judged by a verify-pack; only verifier-green shards ship, failed shards requeue until clean. Five presets, a tiered provider pool, a live scoreboard.
  • Code is one verifiable artifact among many — reports · matrices · audits · benchmarks · dossiers, each claim re-checked in the sandbox (nilcore report renders the trust story and refuses green over a broken chain).
  • Sees the running app through a sandboxed headless browser (drives a flow, then folds a behavioral check into the verdict).
  • Understands four languages structurally and semantically (Go · Python · TS/JS · Rust; AST · graph · repo-map · LSP · pure-Go HNSW · impact).
  • Runs unattended and reacts — metered budgets, durable crash-resume, webhook + cron triggers, gated draft PRs (the agent never merges).
  • Conversational front door (nilcore chat, Telegram/Slack) with live streaming + queue/steer.

The receipts

  • 7 / 7 invariants hold · 2 core deps in the default binary (pure-Go SQLite + golang.org/x/sys) · CGO_ENABLED=0 · the Charm TUI links only under make tui.
  • make verify green · golangci-lint 0 issues · ~46.5k lines of agent (~89.3k incl. tests, 206 test files) · 84 packages.
  • Cross-compiled binaries attached for darwin/linux × amd64/arm64.

What's next

The only remaining roadmap is the gated external-infrastructure tier (EXT-01..08) — deliberately not eligible work, fully blueprinted in docs/EXT-EXECUTION-PLANS.md but blocked behind a recorded human thesis decision. Phase-13+ candidates are scanned in docs/HORIZON.md.

Full changelog: CHANGELOG.md.

v0.1.0

Choose a tag to compare

@github-actions github-actions released this 14 Jun 20:17