Skip to content

Release v0.5.0 - #42

Merged
24601 merged 2 commits into
mainfrom
cursor/release-v0.5.0-34c7
Sep 20, 2026
Merged

24601 merged 2 commits into
mainfrom
cursor/release-v0.5.0-34c7

Conversation

@24601

@24601 24601 commented Sep 20, 2026

Copy link
Copy Markdown
Owner

Prepare the v0.5.0 cut from latest main after merged #40 (llm-to-jev §118) and merged #39 (0843 hygiene, 38e4e92). Rebased onto that tip after #39 landed. Does not reopen or amend open #41 (0947 fold).

Does not create the GitHub Release or tag. After merge, parent cuts:

gh release create v0.5.0 \
  --repo 24601/Augustus \
  --title "v0.5.0" \
  --notes-file docs/release-notes-v0.5.0.md

Head SHA: d28efa25bcbd4bd6dd317978a66029a8ddd8c483 (rebased onto 38e4e92)

Peg: typesafe-ai/skills v0.5.7 (65a39f3). Live HEAD of that repo is still this commit, the only tagged official-skill revision.

High-level summary

Nine commits after the v0.4.0 tag, grouped (not every hourly §):

  • Pages / onboarding. Custom site layout. With vs without Augustus comparison on README and Pages (docs/assets/with-without-augustus.svg).
  • README. Scannable skill map. No uniqueness wall.
  • Recipes (class, not Jev-only). Problem → without → with → what to measure, for GLiNER/GLiClass, Laya, SemIf, NanoJev, kev, Jeff-1, localjev, llm-to-jev, and jcr. Same call-then-act vs place-then-judge split. Shape of improvement, not invented scores.
  • Class / migration. llm-to-jev conversion on-ramp; SemIf rename densify; NanoJev unified-games densify; jcr capability resolver (docs ≠ execute).
  • Measurement honesty. 0843 hysteresis / instruct-tuning ECE / hop-ECE (including Post-#35 hygiene: §114 citation, scannable dumps, hop-ECE interleave #39 even/odd interleave). 0743 Harbor-jevals / Verdict / DecisionOps as themes. Ranking ≠ calibration. Soft Noul ≠ hard gate.
  • CHANGELOG Unreleased dump folded into [0.5.0]. Verbose hourly locks stay in research/changelog-hourly.md.

No invented metrics. No soft-Noul-as-hard-gate. No pipeline() install recipes.

Version pins

  • .claude-plugin/marketplace.json 0.4.00.5.0
  • .agents/skills/augustus/SKILL.md YAML metadata.version 0.4.00.5.0
  • CITATION.cff 0.5.0
  • README / Pages kicker / SECURITY supported line 0.5.x

Discoverability (in-repo only)

  • README: badges, homepage, Install (marketplace / skills.sh / copy path), Not a TypeSafe product, rh-guard, recipes next to the with/without diagram.
  • docs/index.md: v0.5.0 kicker, comparison, #recipes cards, install, pillars. Pages gate strings preserved.
  • Already listed on awesomejev.com and several awesome-jev catalogs. This PR does not spam those lists.
  • OpenSSF Scorecard workflow already on main. Do not invent a numeric score.

Topics for parent (gh repo edit) — 20-cap, do not break

Current topics (20/20): agent-skills agent-workflows agentic-ai ai-agents calibrated-confidence claude-code claude-code-skills codex-skills cursor-skills decision-systems decision-theory formal-methods jev reranking structured-output system-one system-one-models typesafe typesafe-ai zero-shot-classification

README still lists extras that cannot land without a swap: decision-making, mixed-architecture, calibration, gliclass, open-weights, value-of-information. Optional parent swaps only. Community health files already present. has_discussions is still off (optional). Do not invent stars.

Adversarial test results — GATE PASS

Ran after rebase onto merged #39 (38e4e92). GitHub Check Pages site / build is green on run 35524107053.

Check Result
Rebase onto origin/main (38e4e92) CLEAN / MERGEABLE
python3 .agents/skills/augustus/scripts/evaluate_decisions.py --self-test self-test ok (local scoring math; hop-ECE reverse + even/odd interleave; no model call)
python3 .agents/skills/augustus/scripts/uniqueness_gate.py uniqueness-gate ok (21 overlays; 0843/0915/jcr/0922/0940 consecutive)
YAML safe_load SKILL frontmatter name=augustus license=MIT version=0.5.0 typesafe_skill=v0.5.7 @ 65a39f3 description_chars=52646
Version pins marketplace / CITATION.cff / README / Pages kicker = 0.5.0
Pages gate strings TypeSafe Jev Choice/Score/Noul, Install the skill, LICENSE present; ecosystem.md present
GitHub Pages workflow SUCCESS on rebased head
Recipe surfaces README recipes + Pages #recipes + docs/release-notes-v0.5.0.md cards; class not Jev-only
Release-notes honesty ranking ≠ calibration; soft Noul ≠ hard gate; heuristic conversion ≠ calibrated Noul; no invented Scorecard number
Secret scan no new .env / *.pem / assigned API_KEY= literals in this diff

scripts/hourly-refresh.sh was not run (it commits and pushes to main). Did not add npm/pip dependencies.

Out of scope

Rebase note

#39 merged while this PR was open (mergeable_state: dirty). Rebased onto 38e4e92. Kept the scannable [0.5.0] fold and recipe cards. Did not paste the #39 Unreleased dump wall back into CHANGELOG. Noted #39 hygiene in Changed (hop-ECE interleave, uniqueness_gate dump-wall / Pages LICENSE, ecosystem §114 cite). Open #41 remains the only in-flight fold.

Open in Web Open in Cursor 

cursoragent and others added 2 commits September 20, 2026 16:52
Bump marketplace and SKILL YAML to 0.5.0. Fold post-0.4.0 work into a
Keep-a-Changelog cut with class-wide recipes (encoder, open heads,
NanoJev, llm-to-jev, jcr, localjev). Pages and README point at the
with/without diagram. Hourly uniqueness stays in changelog-hourly.md.

Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>
Make ranking ≠ calibration, soft Noul ≠ hard gate, and heuristic
conversion ≠ calibrated Noul greppable in the notes without changing
the meaning.

Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>
@cursor
cursor Bot force-pushed the cursor/release-v0.5.0-34c7 branch from 798699b to d28efa2 Compare September 20, 2026 16:53
@cursor

cursor Bot commented Sep 20, 2026

Copy link
Copy Markdown

Testing gate: PASS

Adversarial review + tests after #39 merged (38e4e92). This PR was dirty vs main; rebased onto that tip. Recipes and the scannable [0.5.0] CHANGELOG fold were kept. Did not merge.

Head: d28efa25bcbd4bd6dd317978a66029a8ddd8c483
Base: 38e4e92 (main)
GitHub: MERGEABLE / CLEAN. Pages: run 35524107053 SUCCESS.

# Check Result Evidence
1 Version bump 0.5.0 PASS .claude-plugin/marketplace.json 0.5.0; SKILL YAML metadata.version: 0.5.0; CITATION.cff 0.5.0; README Current: **0.5.0**; Pages kicker v0.5.0, System One; SECURITY.md supported 0.5.x. SKILL frontmatter yaml.safe_load: name=augustus license=MIT typesafe_skill=v0.5.7 @ 65a39f3 description_chars=52646.
2 CHANGELOG scannable; class recipes PASS [0.5.0] is 79 lines. [Unreleased] empty. Zero uniqueness-lock dumps in CHANGELOG (they stay in research/changelog-hourly.md). Recipes: encoder (GLiNER/GLiClass), open heads (Laya, SemIf, kev, Jeff-1), NanoJev, llm-to-jev, jcr, localjev. Not Jev-only.
3 Release notes honest PASS docs/release-notes-v0.5.0.md cards present. “Numbers below are shapes, not new benches.” Greppable: ranking ≠ calibration; soft Noul ≠ hard gate; heuristic conversion ≠ calibrated Noul. No invented Scorecard number. No pipeline() install recipes.
4 Pages comparison + recipes; uniqueness_gate PASS docs/index.md includes comparison.html and recipes.html; hero CTA #recipes; comparison caption links #recipes. python3 .agents/skills/augustus/scripts/uniqueness_gate.pyuniqueness-gate ok (21 overlays; 0843/0915/jcr/0922/0940 consecutive). Pages strings in docs/index.md + LICENSE in docs/_layouts/default.html.
5 Repo tests PASS python3 .agents/skills/augustus/scripts/evaluate_decisions.py --self-testself-test ok (includes hop-ECE reverse and even/odd interleave from #39). No other test harness (tests/, pytest, npm). Pages workflow SUCCESS on this head. hourly-refresh.sh not run.
6 Secrets / SECURITY / CONTRIBUTING PASS Diff has no .env / *.pem / assigned API_KEY= literals. SECURITY.md still: no hosted API, GitHub Advisories, Dependabot + secret scanning + push protection. CONTRIBUTING.md: merged folds #23–#40, in-flight open #41, hourly dump pointer, local self-test + uniqueness_gate.
7 Independence from #39/#41 PASS (post-rebase) #39 is merged and is now ancestry, not this diff. This PR vs main does not touch uniqueness_gate.py / evaluate_decisions.py / notes.md. Open #41 (0947) is still open and CONFLICTING on its own branch; this diff has no 0947 / jev-as-judge / OneForward / candidate_mass content (only the changelog sentence that #41 is not in this cut). Shared path names (SKILL, CHANGELOG, README, CONTRIBUTING, changelog-hourly) are version/docs only.
8 Human-facing prose PASS (no rewrite) Neutral commits already on the branch. Did not restyle em-dashes.

Rebase resolution (fixable conflict, now landed)

Inherited, not a #42 defect: uniqueness lock UNIQ_0922 still contains do not push onto open #39/#40 as a frozen consecutive substring on main. Changing it would rewrite 21 overlays. Out of scope here.

Parent: merge when ready. Do not cut the GitHub Release from this comment.

@24601
24601 merged commit fb15455 into main Sep 20, 2026
1 check passed
@24601
24601 deleted the cursor/release-v0.5.0-34c7 branch September 20, 2026 16:55
This was referenced Sep 20, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants