Changelog
Every release, pulled live from GitHub.
v1.35.0 — Wave 54 CCA-F Audit Harness Overnight (fable-observe extension) · 2026-06-10
Mooter executes CCA-F-style synthetic audit overnight on local Ollama + Sonnet self-judge. Honest disclaimer: NOT official Anthropic exam.
- Phase 0 Recon — 4 refutations to Cowork brief V2 (paths real are fable-observe not cca-f; Pastor LoRA != tier; judge transport abstraction; Phase C learning staged-only)
- Phase A — Question generator deterministic (60q, seed-reproducible: 16 agentic + 12 cc_config + 12 prompt_eng + 11 mcp + 9 context)
- Phase B — Harness orchestrator (classify.js + Pastor + Ollama + Sonnet self-judge)
- Phase C — Learning loop (samples staged-only, NO retrain, NO fabricated delta)
- Phase D — Report generator (markdown + JSON, honest caveats)
- Phase E — Publish artefacts (notion-page.md + chip.json)
- Phase F.1 — Final-reviewer Opus SHIP-WITH-NITS → nit MED corrected (classify_intact now MEASURED, not asserted)
- Phase F.2 — Selective commit + push + PR
v1.34.0 — Wave 53 Mooter as Local CC Mirror + Anthropic Pride Layer · 2026-06-10
Mooter == Local Claude Code mirror + parallel co-worker. Doctrine V4 enforced at scale.
- A' — Cross-session visibility (enhance existing worktree-conductor/heartbeat.ts, NOT duplicate)
- B — Statusline UX 2026 polish (burn rate + agent focus + ctx fullness + pluggable segments)
- C' — Bash tokens-only extension (post_tool_badge.js, NO settings.json touch)
- D — EMOJI_GUIDE.md + emoji-strict linter (greenfield, anti-hype enforcement)
- E' — Skills aditivas moo-agents/moo-memory/moo-init (8→16, no nativos sombreados)
- G — ANTHROPIC_ALIGNMENT_V2.md + REFUTATIONS_LOG.md (R1-R5 recorded)
- H — 🧪 bench ? honest fallback (anti-fabrication when RESULTS.json absent)
- I' — fable-observe extension cca-f-export.ts (NOT new cca-f tree)
Wave Mega + Wave 52 — OTLP + MooterBench + MCP 16→20 + Foundation + Honest Quota + Fable Observation + Self-Evolving · 2026-06-10
🤖 Generated via Cowork Chrome MCP coordination + CC Wave 52 cherry-pick + Doctrine V4 refutation cycle
- Mooter observability: zero-dep OTLP/HTTP span export of routing decisions
- packages/mooter-bench (Apache 2.0, à la Anthropic Bloom): 50 workflows, 60% accuracy, 100% completion, 62.4% est. savings
- MCP server 16→20 tools (route_query/get_savings/explain_tier/session_summary) + registry submission doc
- 9 + 15 + 27 tests pass
- Advisory cascade with why-not-fable explanation
- span_id feedback loop
- /mooter slash skill
- Security tier-ladder verify
v1.27.0 Anthropic-Aligned · 2026-06-10
Anthropic-aligned honesty layer + Tier 5 Fable (opt-in).
v1.25.0 Statusline Honest · 2026-06-09
Statusline UX overhaul addressing 8 of Paulo's 11 considerations. CC autonomous Day 0 recon refuted multiple kickoff premises (pricing was factually wrong, statusline lives in tools/router/statusline-
- 📚 ctx chip: shows current context window size + percentage
- 🧬 embed nomic 768d chip: clarified as Pastor embedding model
- 🐮 local routes chip: rebrand of MLWR for clarity
- 📊 cost-cap OK chip: real quota numbers (replaced vague limits OK)
- 📝 $X this turn + $Y all-time chips: prompt/session relabeled with token counts
- 🐮 agents N active · M spawned · peak K chip: rename of cryptic 0/0/peak0
- ☁️ Claude Max [███████░░░] N% left chip: visual progress bar
- ⏭️ (deferred) Terminal name auto-refresh — documented in repro file
v1.24.0 Friends Activation · 2026-06-09
Four-phase mega aimed at the first-impression points for incoming friends. CC autonomous execution refuted 3 of 4 kickoff premises via Day 0 recon (honest > forced).
- Premise refuted: the command already existed from Wave 33.5.
- Added friend-facing PT-PT rules: packs, savings→dashboard, explain/router-debug, doctor.
- Friends can now type natural questions instead of memorizing commands.
- Premise refuted: dashboard is NOT minimalist. Already 2303 lines / 6 tabs.
- Single real gap is a Recharts MLWR trend chart, which violates the dependency guardrail of this wave’s scope.
- No-op rather than fake work.
- Premise partially refuted: the page existed but was hardcoded/stale (still showing v1.21.1).
- Rebuilt with ISR (revalidate: 3600) fetching GitHub Releases API publicly, filtered to v1.x.y tags.
v1.23.0 Bug Fix Trinity · 2026-06-09
Fixes 3 bugs surfaced by parallel-CC validation dogfood in Wave 33-VAL.
- buildDigest() now folds subagent dispatches (herd state from SubagentStop hook) into the tier mix.
- prompts vs delegated tracked separately; savings $ stays tracker-driven → not inflatable.
- Smoke: 0 prompts + 6 delegated → T0 67% · T1 33% correctly reflected.
- describeEngineError() now emits actionable message: names the missing dep + suggests npm install command or source-checkout guide.
- No more cryptic "engine unavailable" wall.
- Day 0 recon refuted the kickoff premise. The "$0 saved" symptom was Bug C (subagent dispatches invisible to tracker), not a real ultracode bypass.
- A hook cannot reliably detect the thinking-flag of the CC harness nor force delegation.
- Honest fix: caveat in mooter explain saved instead of building a false detector.
v1.21.11 OAuth Polish · 2026-06-09
Loading state + reason-aware errors for GitHub OAuth sign-in.
- landing/app/auth/callback/route.ts — reason-aware error handling on OAuth callback
- landing/app/_components/AuthErrorBanner.tsx — new component for displaying auth errors
- landing/app/(app)/layout.tsx — layout updates for loading state
- landing/app/(app)/wave44-oauth-polish.test.ts — tests for the new flow
- classify.js sha 7b01eb86… INTACT (23 waves consecutivas)
- packages Wave 28-33.x UNTOUCHED (landing-only change)
- final-reviewer Opus: SHIP 0-HIGH/0-MED
- 139/139 landing tests pass + next build clean
v1.21.8 LoRA Deps Unblock · 2026-06-09
Honest follow-up to the LoRA training setup.
- unsloth==2026.6.1 is the current latest stable
- transformers==4.56.0 is compatible (outside the prior exclusion range)
- pins verified via PyPI metadata
- docs/strategy/LORA_TRAINING_RUNBOOK.md — operational runbook for overnight RTX 4090 training runs (pre-flight checks, dataset path, exact command).
- classify.js sha 7b01eb86… INTACT (22nd consecutive wave)
- packages Wave 28-33.x UNTOUCHED
- version.json zero-touch
- final-reviewer Opus: SHIP 0 nits
v1.21.7 Quiet Cleanup · 2026-06-09
Polish pass on mooter dogfood log UX.
- mooter dogfood log now warns on unknown severity values
- mooter dogfood (no args) hints available subcommands
- +3 tests covering severity validation + subcommand hints
- classify.js sha 7b01eb86… INTACT (21st consecutive wave)
- packages Wave 28-33.x UNTOUCHED
- version.json zero-touch
- final-reviewer Opus: SHIP 0H/0M/1L (cosmetic)
v1.5.0-arch-hardening — Wave 10 Phase C.1 · 2026-06-02
Merge commit: dbe31b6
- F-1 🔴 Hub ingestion rate-limit (NON-BREAKING, no auth)
- /api/delta: per profile_hash (30/60s) · fail-open · 429
- /api/device-heartbeat: per device_id (10/60s) · fail-open · 429
- Existing installs + public install.sh keep working
- Per-IP dimension deferred → F-1.2 (needs KV namespace)
- F-2 🟠 CI P11 gate (classify.js.sha256 lockfile + checksum step)
- F-3 🟠 Landing vitest + typecheck + lint in CI
- F-10 Tooling fix (replay.js IIFE slice) — gold-labels green again