v0.7.0

What's Changed

• deps: pin seven transitive deps for the October-2026 CVE batch by @szjanikowski in #84
• pricing: refresh the catalog as of 2026-10-14 — Claude 5.5 + 4.x legacy, GPT-5.6 + GPT-6 by @szjanikowski in #85
• chore: move default models to Claude 5.5, show a real result in README by @szjanikowski in #86
• chore: release v0.7.0 by @szjanikowski in #87

Full Changelog: v0.6.0...v0.7.0

v0.6.0

What's Changed

• [codex] Apply NASDE branding by @szjanikowski in #73
• ci: scope Quality Gate concurrency to the calling workflow by @szjanikowski in #74
• chore: bump harbor 0.13 -> 0.19 by @szjanikowski in #76
• Calibration round 2: rubric v2, agent-diff judge input, per-task dimensions, judge-matrix run plan by @szjanikowski in #75
• Opus 5: pricing entry + charts with the Opus 5 vanilla arm by @szjanikowski in #77
• deps: pin aiohttp, cryptography and h2 for the August-2026 CVE batch by @szjanikowski in #81
• Heatmap variant with the T2 bounty row outlined by @szjanikowski in #78
• Claude Fable 5.1: pricing entry with the 0.025x cache-read exception by @szjanikowski in #80
• deps: raise the typer floor to >=0.26 (fresh installs died on typer._click) by @szjanikowski in #82
• chore: release v0.6.0 by @szjanikowski in #83

Full Changelog: v0.5.0...v0.6.0

v0.5.0

What's Changed

• feat: variant task-scoping + tactical-ddd skill benchmark example by @szjanikowski in #54
• docs: update tactical-ddd example results to final n=5 by @szjanikowski in #55
• chore(deps): bump harbor 0.6 → 0.13 by @szjanikowski in #56
• docs: link DDD deep-dive to the full blog write-up by @szjanikowski in #58
• feat: results-export + fix repeated-evaluation accumulation by @szjanikowski in #57
• feat(calibrate): rubric calibration via PR/MR review (ADR-010) by @szjanikowski in #59
• feat(metrics): token & cost efficiency per trial (ADR-011) by @szjanikowski in #60
• fix(codex): restore ChatGPT OAuth auth under harbor 0.13 by @szjanikowski in #61
• fix(runner): resolve own job_dir from config, not globally-newest dir (BUG #2) by @szjanikowski in #62
• feat(metrics): raw+std economics, reasoning effort; drop scalar efficiency (ADR-011) by @szjanikowski in #63
• fix(skills): register Codex/Gemini skills natively, not in /app cwd (ADR-012) by @szjanikowski in #65
• docs(adr-012): sync with merged code — per-agent keying + Gemini e2e by @szjanikowski in #66
• fix(skills): route [[skill]] + plugin skills natively for Codex/Gemini (ADR-012) by @szjanikowski in #67
• docs: migrate README into Starlight docs site (website/) on GitHub Pages by @szjanikowski in #64
• docs: reposition messaging — whole-harness cost & migration decision (CTO audience) by @szjanikowski in #68
• chore(deps): pin transitive deps to fix June-2026 CVE batch by @szjanikowski in #70
• docs: rebrand NASDE → Nasde casing across docs site by @szjanikowski in #69
• feat(pricing): convention-based layered price override (ADR-013) by @szjanikowski in #71
• chore: release v0.5.0 by @szjanikowski in #72

Full Changelog: v0.4.0...v0.5.0

v0.4.0

What's Changed

• feat(plugin): [nasde.plugin] + skill-by-reference ([[skill]]) by @szjanikowski in #51
• ci: ignore disputed pyjwt CVE PYSEC-2025-183 in quality gate by @szjanikowski in #52
• chore: release v0.4.0 by @szjanikowski in #53

Full Changelog: v0.3.3...v0.4.0

v0.3.3

What's Changed

• docs(readme): migrate install instructions to PyPI by @szjanikowski in #41
• fix(install): cap requires-python<3.14, add Windows smoke matrix, fix path bug by @szjanikowski in #43
• ci: extend quality-gate matrix to Windows; fix codex backend test isolation by @szjanikowski in #44
• chore(scripts): add Windows PowerShell OAuth token exporter by @szjanikowski in #42
• feat(skill): bundle OAuth scripts into nasde-benchmark-runner by @szjanikowski in #45
• fix(windows): enforce LF line endings for shell scripts and Dockerfile by @szjanikowski in #47
• fix(deps): ship harbor[cloud] extras by default by @szjanikowski in #48
• chore: release v0.3.3 by @szjanikowski in #49
• docs(skill): clarify --all-variants is sequential, not parallel by @szjanikowski in #50

Full Changelog: v0.3.2...v0.3.3

v0.3.2

What's Changed

• feat: PyPI publishing + in-CLI update notifier by @szjanikowski in #36
• fix(build): drop local-version segment from hatch-vcs output by @szjanikowski in #37
• chore(release): operationalize PyPI publishing — bump harbor, smoke gates, docs by @szjanikowski in #38
• chore: release v0.3.2 by @szjanikowski in #40

Full Changelog: v0.3.0...v0.3.2

v0.3.0 — single-file task config + per-dimension scoring

Changed

• Breaking: per-task config consolidated into a single task.toml. The previous split between task.json (nasde-only) and task.toml (Harbor) is gone — Harbor sections ([task], [agent], [environment], [verifier], [metadata]) and nasde-specific fields ([nasde.source]) now live side by side in one file. All bundled examples migrated. (#29)
• Breaking: model is now required in every variant.toml. There is no project-level model default in nasde.toml anymore — different agent families need different models, so each variant declares its own. Missing model fails fast with a clear error. CLI --model still overrides. (#29)

Added

• Independent per-dimension scoring scales (ADR-008). Each dimension in assessment_dimensions.json declares its own max_score (any positive integer). normalized_score is computed from the actual sum of per-dimension maxima — pick the granularity that matches what you can distinguish, instead of forcing every benchmark into a fixed total. (#34)

Fixed

• Evaluator now prints a clear, actionable error when the underlying claude or codex CLI is missing, instead of failing with a confusing subprocess trace. (#30)
• The "missing CLI" hint now prints [evaluation] correctly instead of hiding it as Rich markup.

Security

• Pin litellm>=1.83.7 to address GHSA-xqmj-j6mv-4862. (#31)

Docs

• README: clarified how NASDE is meant to be used end-to-end.
• README: Harbor links now point at harborframework.com instead of the GitHub repo, matching Harbor's primary site.

Internal

• CODEOWNERS added so PRs auto-request review from the right people.

---

Install: uv tool install git+https://github.com/NoesisVision/nasde-toolkit.git@v0.3.0

Full changelog: v0.2.1...v0.3.0

v0.2.1 — Dependency modernization + release governance

Changed

• Modernized dependencies: harbor 0.1.45 → 0.4.0, opik 1.10.39 → 2.0.9.
  Brings upstream bugfixes and removes a long-standing Opik monkey-patch that
  is no longer needed. ConfigurableCodex simplified by ~80 LOC thanks to
  Harbor 0.4 adding first-class support for files-to-inject. (#25)
• Default reviewer model bumped to claude-opus-4-7 across README, ARCHITECTURE,
  bundled nasde-benchmark-runner skill, and example benchmarks
  (ddd-architectural-challenges, refactoring-skill). nasde-dev-skill
  stays on Sonnet 4.6 as documented in benchmark results.

Security

• Added pip-audit as a CI gate. Fixes 20 known CVEs across the transitive
  dependency tree, including 1 CRITICAL and 3 HIGH in litellm / cbor2.
  New vulnerabilities will now fail the quality gate instead of shipping
  silently. (#25)

Added

• CHANGELOG.md, SECURITY.md, docs/RELEASING.md. (#26)

Docs

• README: unified "what NASDE does" opener around four steps, mentioning the
  rough test.sh pass/fail gate that precedes LLM-as-a-judge scoring.
• README: scoring section now calls out that criteria can cover the agent's
  run-time signals — tool-call trajectory, token usage, wall-clock duration —
  alongside the produced artifacts.

v0.2.0 — Subprocess evaluator + bundled skills

Highlights

• Subprocess evaluator backends (#24). LLM-as-a-judge reviewer now runs as a subprocess against your installed claude or codex CLI instead of going through the Anthropic Agent SDK. Select with [evaluation] backend in nasde.toml. Unlocks subscription billing (Claude Max, ChatGPT Plus) for local evaluation and reuses your existing OAuth/keychain auth.
• Bundled authoring skills + nasde install-skills. Copies the nasde-benchmark-* skills into ~/.claude/skills/ so they're available in every Claude Code session. --scope project and --force flags supported.
• Trajectory-aware evaluation (#22). Set [evaluation] include_trajectory = true to let the reviewer access the agent's ATIF trajectory (agent/trajectory.json) alongside the final workspace — criteria can now reward how the agent got to the result.
• MCP server configuration per variant. Drop a claude_config.json into a variant directory to wire up MCP servers for that variant's runs.
• README rewritten for a developer audience with problem-first framing and an explicit "why NASDE" section.

Fixes

• Evaluator captures a real stderr file descriptor, surfacing actual error output when the subprocess fails.
• Clearer diagnostics when the evaluator's Claude Code process exits non-zero.
• scripts/export_*_oauth_token.sh no longer use set -e, so sourcing them into a shell with an error no longer kills the terminal.
• nasde eval now forwards the full [evaluation] config from nasde.toml (#23).

See CHANGELOG.md for full details.

v0.1.1 — hatch-vcs versioning

Versioning and docs polish on top of v0.1.0.

Changes:

• Version from git tags via hatch-vcs — no more hand-edited version = "…" in pyproject.toml. _version.py is auto-generated and gitignored.
• README badges (Noesis Vision, CI status, License) and stable-vs-HEAD installation instructions.
• _version.py excluded from ruff and mypy.
• Removed auto-generated harbor_config.json from shipped examples (was leaking per-machine absolute paths).

See CHANGELOG.md for full details.