v3.19.0 — Asolaria Absorption: watcher-gate, BH tracing, claims, handoff

Absorção do ecossistema Asolaria/Jesse\n\n### Watcher-gate pre-promise:\n- Verificação watcher antes do <promise>\n- Gate: MEASURED (pass) / UNVERIFIED (fail)\n- Integrado nos 3 mirrors (hooks/, plugin/, _bundle/)\n\n### Brown-Hilbert tracing:\n- bh_address() para endereçamento port.port.port\n- cmd_delegation() — reconstroi árvore de delegação\n\n### Claims-gate:\n- MEASURED/UNVERIFIED tagging em todo output\n- 8 regras Asolaria\n- claims-gate --check command\n\n### Agent-to-agent handoff:\n- Spindle/latch protocol em handoff.py\n- handoff --next <agent> --state <json>\n- Integração com simplicio CLI

v3.18.1

Fixed

• references/orchestration.md Step 3c still said "TRIVIAL/SMALL skip adversarial review" — a
  leftover contradiction from before v3.17.0's 6-agent floor, found by an independent audit. Now
  consistent: the review fan-out runs on every item; only Rubric C's heaviest sub-checks stay
  tier/surface-scoped.
• Deleted .agents/skills/, an untracked gitignored local mirror that was a week stale.

Verification: python3 scripts/check.py — claims-audit 5/5, 44 tests passed.

v3.18.0

Changed

• Native simplicio-runtime bind is now REQUIRED, not optional, on 8 of the 11 supported
  runtimes (Claude Code, Codex, Cursor, VS Code, Antigravity, Kiro, OpenCode, Hermes). The
  installer (ensure_runtime_bind()) now applies the bind unconditionally for these hosts
  instead of printing a suggestion: simplicio install --global (Claude/Codex/Cursor/VS
  Code/Kiro in one pass), fixed OpenCode auto-registration, and a simplicio doctor --json
  health check for Hermes/Antigravity. Every generated entry file states the requirement inline.
• Fixed a broken MCP invocation referenced across 5 adapter READMEs and OpenCode's
  registration code — simplicio-cli mcp register --client <x> isn't a real registration path;
  the correct command is simplicio install --global / simplicio serve --mcp --stdio. Also
  fixed the args order (["mcp","serve"] → ["serve","--mcp","--stdio"]) in 4 README snippets.
• Gemini, Aider, and OpenClaw are unaffected — bind stays optional/native-by-design there.

Verification: python3 scripts/check.py — claims-audit 5/5, 44 tests passed.

v3.17.0

Changed

• Minimum 6-agent orchestration floor (simplicio-tasks/simplicio-review). Removed the
  solo/inline fast-path and the TRIVIAL/SMALL single-self-review shortcut. Every item, regardless
  of size, now runs a 6-role floor: orient+plan, implement, 3 parallel adversarial review rubrics
  (security/correctness, quality, does-it-reproduce — simplicio-review), and an independent
  blast-radius reviewer that re-checks impact_audit.json/flow_audit.json instead of trusting
  the implementer's self-reported scope. The Step 3 auto-scaling formula gained a max(6, ...)
  floor; when disk is too tight for 6 worktrees, the floor still runs under isolation=shared
  instead of dropping below 6. Rubric C (does-it-reproduce) is no longer gated to LARGE/CRITICAL —
  it runs always; only its heaviest sub-checks (web/flow evidence) stay scoped to diffs that touch
  that surface.

Verification: python3 scripts/check.py — claims-audit 5/5, 44 tests passed.

v3.16.0

Added

• Cross-agent handoff on incomplete stop (hooks/loop_stop.py). A budget-halted or
  iteration-cap stop now writes .orchestrator/loop/HANDOFF.md before clearing the scratchpad —
  the frozen goal/acceptance criteria from task_anchor.py, the last attempts from
  loop_journal.py, and concrete resume steps. Previously these stops deleted the scratchpad with
  nothing durable left for a different agent/runtime to pick the task back up cold; a successful
  (promise-fulfilled) stop is unaffected and still needs no handoff.

Verification: python3 scripts/check.py — claims-audit 5/5, 44 tests passed.

v3.15.0

Added

• add scripts/impact_audit.py to force scope reflection across callers, local dependencies, and related tests before editing
• add scripts/flow_audit.py to map UI actions, frontend calls, backend endpoints, and service calls in mixed front/back/service workspaces

Changed

• require impact-audit and flow-audit evidence in simplicio-tasks, simplicio-loop, and simplicio-review
• bump simplicio-loop to 3.15.0 and align the npm wrapper with the new PyPI version

v3.14.1

Changed

• require simplicio-mapper>=0.11.0 and simplicio-cli>=0.7.1 in package metadata
• make fresh installs deterministic before the loop preflight auto-update runs

Verification

• python3 scripts/check.py
• python3 -m build
• python3 -m twine check dist/*

v3.14.0 — auto-update operators at preflight

Changed

• The loop preflight now auto-updates the operators to their latest release. Before iteration 1, simplicio-loop runs python3 -m pip install -qU simplicio-mapper simplicio-cli (PEP-668 --user fallback) so every run uses the newest mapper/dev-cli — no manual re-install needed to pick up a new simplicio-mapper. Fail-open and offline-safe; runs once per preflight, not per turn.

Verification: python3 scripts/check.py — claims-audit 5/5, 35 tests passed.

v3.13.0 — mapper two-tier async survey

Changed

• Wired the simplicio-mapper two-tier async survey (v0.9+) into the loop. The orient / survey step now prefers simplicio-mapper scan . --json (instant macro skeleton + deep index in the background) over the blocking index crawl, polls simplicio-mapper status . --json for the deep-pass phase, and uses macro . --json for instant triage. index . --json is retained as the forced-synchronous build (scan --sync/--await for blocking waits). Updated simplicio-loop/SKILL.md, simplicio-tasks/references/extension-points.md, and the understand-anything adapter; mirrored into plugin/ and _bundle/.

Verification: python3 scripts/check.py — claims-audit PASS (5/5), 35 tests passed.

v3.11.0 — remove headroom + all Rust references

Removed

• The vendored native performance core and all its modules (simplicio-core / -py / -proxy / -parity) — removed in full, along with their build/parity tooling.
• The ported ONNX model engine modules and their subcommands — kompress, router, embed, and image, plus the GitHub Copilot OAuth port (copilot).
• The [onnx] and [kompress] optional-dependency extras and the ONNX-model doctor checks (onnxruntime / huggingface_hub / tokenizers / pillow no longer pulled in or probed).
• All references to and integration with the external compression tool. The native, stdlib-only Simplicio capture engine (deterministic compression) is now the sole engine.
• Rust as a referenced target stack. Dropped all Rust/ecosystem references project-wide — language-detection rule, .rs signature mapping, cargo* clamp allow-list entries, and every cargo/crate example across the README + 13 translations, the skills, and the docs. The separate external simplicio-runtime is unaffected.