Skip to content

v0.4.1 — the onboarding overhaul

Choose a tag to compare

View all tags
@matt82198 matt82198 released this 26 Jul 18:18
v0.4.1
e460fd6
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Patch release: the adversarial onboarding loop ran to zero verified P0/P1 — this ships it, plus the credibility hardening that followed an external review.

Onboarding overhaul

  • Scaffolder: absolute aesop_root, real NEXT-STEPS guidance (actual port + absolute paths) (#407, #413)
  • aesop doctor: config-structure validation, placeholder-URL warning, skills-copy check, Node/Python version checks (#407)
  • Setup docs truth-sync: CONFIGURE field names match generated output, Claude Code install steps, env-var reference, backlog format (#405)
  • run-watchdog.sh --once now ends with an explicit WATCHDOG SMOKE: PASSED/FAILED verdict + honest exit codes (#406)
  • reproduce degrades gracefully in fresh scaffolds; config ~ expansion (#413)

Dashboard

  • Cost view: per-wave / per-agent / per-model spend breakdown with expandable model mix (#402)

CI & test hardening

  • PowerShell syntax gate (#401) · workflow_dispatch on main-full (#399) · self-updating test-count gate --fix (#403) · frontier-slice test depollution (#400)
  • Git-identity poisoning root-caused + runtime tripwire (the "Test User" ghost is dead) (#412)
  • Wave-manifest preflight lint (wave_manifest_lint.py) — file-ownership disjointness + prompt sanity before dispatch (#409)
  • Wave quality scorecards from ledger telemetry (wave_scorecard.py, StateAPI-conformant) (#408)

Honesty & measurement

  • Pre-registered equivalence protocol (bench/EQUIVALENCE-MARGIN.md, committed BEFORE results) (#410)
  • Judgment-v3 ceiling addendum: our own pre-declared ceiling rule disqualifies the 39/39 headline as an equivalence claim — it is a sufficiency floor; the frontier discrimination slice exists because the protocol caught our own number (#414)
  • Wave-24 boundary named: cheap models hold on scored bounded judgment, break on open-ended selection — the verification layer is the mitigation (#414)
  • License history disclosed: MIT until 2026-07-17, PolyForm Strict 1.0.0 since (#414)
  • self_stats counts squash merges via gh (268 → 387 merged PRs); single stats source + verify-stats.sh (#398, #411)
  • 4o-mini worker-seat metrics: 5/5 on bounded tasks at $0.00066 total — seam proven under load, wave-scale explicitly unproven (#404)

Full changelog: CHANGELOG.md ([0.4.1] section; late-merging items above are canonical here).

Assets 2
Loading