Reef - Browse /v0.1.0 at SourceForge.net

The interactive file manager requires Javascript. Please enable it or use sftp or scp.
You may still browse the files here.

Name	Modified	Size	InfoDownloads / Week
Parent folder
reef_infra-0.1.0-py3-none-any.whl	2026-09-23	1.1 MB	0
reef_infra-0.1.0.tar.gz	2026-09-23	7.0 MB	0
README.md	2026-09-23	40.2 kB	0
v0.1.0 source code.tar.gz	2026-09-23	6.9 MB	0
v0.1.0 source code.zip	2026-09-23	8.0 MB	0
Totals: 5 Items		23.0 MB	0

What's Changed

ci: set repository for release creation by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/148
feat: add evolution budgets, whole-tree credential scan, and client release identity by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/146
fix(service): parse Bearer auth scheme case-insensitively by @mmaxjr in https://github.com/Human-Agent-Society/reef/pull/150
fix: reapply the theme on client-side navigation, not only full loads by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/149
feat: add a code copy button and align the logo with the sidebar by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/154
fix: repair docs anchor landing, dialog focus, search, and mobile code blocks by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/153
fix: keep fact-table label columns on one line on phones by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/156
fix(service): reject invalid reef serve arguments instead of ignoring them by @mmaxjr in https://github.com/Human-Agent-Society/reef/pull/151
fix: center the code copy button in one-line blocks on phones by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/158
feat: run episodes through a pluggable executor with a bubblewrap sandbox by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/152
docs: link organization discussions by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/162
feat: grow the evolution gate with promoted failing traces by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/155
fix(service): answer 400 for non-object request bodies and non-finite report scores by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/167
fix(deploy): validate the stack config structure and unique service names before startup by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/168
feat: recheck published trees and roll back a regressed publish by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/157
fix: forward episode executor from Cordis recipe by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/174
feat: add a DeepSeek Harness (dsh) adapter by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/173
feat: Implement GEPA in Reef by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/47
fix(deploy): bind Python service commands explicitly by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/176
harness manifest: distinguish "no surface" from "no artifact" in the 404 message by @Xuan-1998 in https://github.com/Human-Agent-Society/reef/pull/179
Expose bundled adapters via GET /reef/adapters by @Xuan-1998 in https://github.com/Human-Agent-Society/reef/pull/180
docs: clarify OpenClawRL session correlation by @noah-ing in https://github.com/Human-Agent-Society/reef/pull/186
Improve a few pages in doc site. by @yanxz in https://github.com/Human-Agent-Society/reef/pull/185
feat: hold gate wins for review as pending releases with a promote route by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/166
fix(deploy): terminate managed process trees by @nevasini1 in https://github.com/Human-Agent-Society/reef/pull/188
fix(cordis): clean rendered artifact sources by @nevasini1 in https://github.com/Human-Agent-Society/reef/pull/190
fix(service): relocate the install wrapper by the env entry above the config target by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/183
feat: add a Hermes Agent (hermes) adapter and a git install kind by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/184
feat: add a native harness whose loop lives in the tree and whose tools are nodes by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/169
[Task] Close five small documentation, tooling, and test gaps by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/195
perf(training): persist the adapter slot without capturing it twice by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/198
perf(training): write the adapter slot snapshot off the paused window by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/200
docs: add a Chinese README by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/194
feat(harness): add the Codex adapter by @simonucl in https://github.com/Human-Agent-Society/reef/pull/193
feat: open three seams of the native loop to native_hook nodes by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/170
docs: add Star History charts to READMEs by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/211
docs: add team sections to English and Chinese READMEs by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/212
fix name by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/213
feat(harness): add a Terminus 2 adapter by @simonucl in https://github.com/Human-Agent-Society/reef/pull/209
docs: catch the reference pages up with native, codex, and terminus by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/216
optimize docs by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/217
docs: add coral favicon by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/224
Cache commit history for constant-time training bookkeeping by @morluto in https://github.com/Human-Agent-Society/reef/pull/221
Preserve record conflicts after compaction by @morluto in https://github.com/Human-Agent-Society/reef/pull/222
Keep a stored report canonical when discarding it by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/231
refactor: use declared interfaces and plain value types by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/234
refactor: stop probing declared interfaces, and split two overloaded reads by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/237
Stabilize flaky CI test in commit-log harness growth path (Python 3.10) by @BobbyZhouZijian with @Copilot in https://github.com/Human-Agent-Society/reef/pull/238
feat(native): spill a tool result over the cap to a workspace file the model can read by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/236
fix: send CLI usage errors to stderr and exit 2 by @ump45nose in https://github.com/Human-Agent-Society/reef/pull/232
fix(openclawrl): bound the session scan and report correlation failures by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/240
Make training and rollback settlement retry-safe by @morluto in https://github.com/Human-Agent-Society/reef/pull/223
fix(records): remove dead receipt conflict check by @Marsh-Yan in https://github.com/Human-Agent-Society/reef/pull/244
Remove the unreachable clause in the compaction-receipt conflict check by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/242
feat(cordis): treat the traffic that drives harness evolution as untrusted input by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/239
docs: clarify page titles and connect learning guides by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/246
feat(tutorial): run harness_evolve on the native harness by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/241
feat(native): add the pre_execute hook event with allow, deny and ask, and tool capabilities by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/243
feat(harness): run a pulled native tree through the client wrapper by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/245
fix(service): reject invalid nested serve overrides by @Marsh-Yan in https://github.com/Human-Agent-Society/reef/pull/249
fix(service): reject empty or whitespace-only release_id on scenario create by @v01dst in https://github.com/Human-Agent-Society/reef/pull/247
feat(native): run the loop from a native_graph node by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/250
feat(docs): lay the site out in two header rows, a full sidebar and a table of contents by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/251
perf(docs): redirect the docs entry at the edge and load the fonts in parallel by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/253
fix(docs): line the sidebar text up with the logo and the tabs by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/252
fix(docs): center the header rows at the layout width so the tabs line up with the sidebar on a wide screen by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/254
docs(harness): add an at a glance summary and the adapter table to the harness evolution guide by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/255
docs: clarify Reef positioning and artifact evaluation by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/256
docs: keep English and Chinese READMEs synchronized by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/257
docs(tutorial): measure the graph node on the native tutorial by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/258
fix(harness): make the README install flow work: seed base artifact and model binding by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/265
feat(native): add branch and compact stages to the loop graph and an execute seed tool by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/259
fix(harness): refuse an update that changes an entry's kind by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/264
feat(native): add native_agent nodes and a subagent stage so a graph can delegate by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/260
feat(native): enforce declared tool capabilities per call under the sandbox and log the enforcement by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/268
fix(harness): make the README harness evolution flow work end to end by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/271
chore(compose): port cordis 4.0.0-rc.9 fixes and pin the conformance map by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/270
fix(native): choose the enforcer before any tree module runs and keep a sandbox failure out of the tool error count by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/276
feat(native): register node entries as effects into a host the interpreter reads at each step by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/272
docs(tutorials): one directory per tutorial with the notebook first and an index by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/261
feat(evolve): persist each evolve step's record so the self modification decision is reconstructible by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/267
fix(artifacts): isolate LFS hooks in managed clones by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/280
feat(native): serve the tree as a resident process with live mounts, a proposals route and the self tools by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/278
feat(harness): install pinned agent binaries at startup by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/273
fix(native): poll again at the interval when a release poll times out behind a running evolve step by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/282
fix(scenario): serve committed releases during long-running evolution by @noah-ing in https://github.com/Human-Agent-Society/reef/pull/285
refactor(harness): lay reef/harness out as tree, adapters, episodes, runners and client by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/283
fix(docker): preserve NumPy pin in tttd solver install by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/288
fix(harness): rank an errored native turn as could not run, bound proposed patterns, route a proposed agent by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/284
fix(tttd): preserve training image PYTHONPATH by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/291
fix(deploy): clear stale bridge readiness before stack startup by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/292
test(harness): restore the four example tests dropped by [#284] by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/293
docs(tutorial): a self tools form and a replay page of what a run did by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/287
chore(harness): remove the forwarding modules kept after the [#283] move by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/302
Fix config propagation for dotted recipe references by @sudhanshuskarn in https://github.com/Human-Agent-Society/reef/pull/295
fix(compose): keep one row per entry when Loader.update moves it by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/298
fix(native): read a tool module's declaration statically and import it only where the call runs by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/299
fix(harness): configure upstream model at startup by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/307
feat(native): add the native_loop kind, the turn loop as reviewed code by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/300
docs(tutorial): call the replay page's tool call table tool calls by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/306
fix(bridge): ship a resolvable reference so external loss families start by @Xuan-1998 in https://github.com/Human-Agent-Society/reef/pull/297
fix(compose): keep a group entry's config as its children's live list by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/309
feat(meta-harness): add full-history harness search by @simonucl in https://github.com/Human-Agent-Society/reef/pull/214
docs: animate README architecture diagram by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/317
feat: add native auto/manual training and mode switching by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/316
chore(deps-dev): bump ruff from 0.14.7 to 0.16.6 in the python-minor-and-patch group by @dependabot[bot] in https://github.com/Human-Agent-Society/reef/pull/321
chore(deps): bump the docs-site-minor-and-patch group in /docs/site with 6 updates by @dependabot[bot] in https://github.com/Human-Agent-Society/reef/pull/322
feat(serve): resolve a dotted recipe's package beside the config by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/323
fix: keep admission checks and manual step guarantees on POST /reef/train by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/319
refactor(runtime): unify worker executors and reuse evolution pools by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/279
feat: add training_mode hybrid so a deployment takes instructions while it learns from failures by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/320
fix(runtime): harden executor cleanup and local Ray startup by @Chonghe-Jiang in https://github.com/Human-Agent-Society/reef/pull/326
Revamp the architecture.rst doc page by @yanxz in https://github.com/Human-Agent-Society/reef/pull/324
feat(harness): submit harness requests through native manual training by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/311
docs: add WeChat group link and QR code page by @quao627 in https://github.com/Human-Agent-Society/reef/pull/327
fix(openclawrl): resume hermes sessions across turns and restart the stack after training by @MikeZheng777 in https://github.com/Human-Agent-Society/reef/pull/329
fix(openclawrl): run the container side of a session as the host user by @MikeZheng777 in https://github.com/Human-Agent-Society/reef/pull/331
[Example] CORAL test-time training: gateway middleware, call journal, reporter, training recipe (issue [#3]) by @Xuan-1998 in https://github.com/Human-Agent-Society/reef/pull/296
fix(train): size LoRA checkpoint reservations from the base model, not 8x by @Xuan-1998 in https://github.com/Human-Agent-Society/reef/pull/328
test: fix pid-marker read race in killed-worker executor test by @Xuan-1998 in https://github.com/Human-Agent-Society/reef/pull/334
fix(dispatcher): log checkpoint-storage block transitions with reasons by @Xuan-1998 in https://github.com/Human-Agent-Society/reef/pull/335
feat(harness): Harness requests stage 2: the tutorial's pi deployment reviews evolved extensions before serving them by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/312
feat(harness): Harness requests stage 3: what a release needs from the person, checked off before it installs by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/313
feat(harness): Harness requests stage 4: a page per step and /reef-versions in reef-pi by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/314
feat(harness): Harness requests stage 5: the two demos as a tutorial, and the measurement stage 6 reads by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/315
fix(sao-example): block until each task's rollouts have trained by @Xuan-1998 in https://github.com/Human-Agent-Society/reef/pull/338
fix(train): carry the driver's PYTHONPATH to the Ray actors its job creates by @Xuan-1998 in https://github.com/Human-Agent-Society/reef/pull/337
fix(tutorial): hand the proposer each failure's report score and feedback by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/340
feat(harness): bind the install script to the forwarded host behind a gateway by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/342
fix(runtime): keep serving alive when adapter capacity refuses a publication by @HectorHHZ in https://github.com/Human-Agent-Society/reef/pull/344
docs(harness): add the first harness requests runs on main to the tutorial README by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/339
fix(deploy): use managed Ray for CORAL GPU placement by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/346
feat(sao): the paper's value-model warmup and learning rate in the example by @Xuan-1998 in https://github.com/Human-Agent-Society/reef/pull/341
docs(runtime): settle the residency window against the staleness bound by @HectorHHZ in https://github.com/Human-Agent-Society/reef/pull/345
feat(harness): the install script says what it is doing by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/347
feat(harness): a spinner while the install script waits on npm and pip by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/348
fix(harness): what a person's reef-pi run needs that an episode does not by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/349
[Example] CORAL TTT: drive the real CORAL runtime in coral_demo by @Xuan-1998 in https://github.com/Human-Agent-Society/reef/pull/336
feat(service): DELETE /reef/scenarios/{scenario} retires a scenario and archives its state by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/350
feat(harness): record the tokens each evolve step's model calls used by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/351
feat(harness): evolution.client_models, the models an installed client may switch to by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/353
docs: add shared agent instructions by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/361
refactor: simplify terminology across code and docs by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/360
feat: resolve harness evolve models per scenario with BYOK by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/362
refactor: replace BYOK resolver with scenario model settings by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/363
feat: connect local Reef runtimes to the dashboard by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/364
feat: retain compacted traces with automatic retention limits by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/366
Expose raw learning history and retain model reasoning traces by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/368
fix(harness): reef-pi runs from any shell by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/372
fix(harness): the install script pins its python3 into the wrapper and keeps cwd off sys.path by @quao627 in https://github.com/Human-Agent-Society/reef/pull/373
feat(artifact): record training metadata with every published LoRA adapter by @HectorHHZ in https://github.com/Human-Agent-Society/reef/pull/369
fix(harness): relink reef-<adapter> on every install run, and give the install tests a home of their own by @quao627 in https://github.com/Human-Agent-Society/reef/pull/386
feat(serve): the harness evolve recipe carries a profile that reef serve --recipe starts by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/388
feat(harness): reef serve prints the install line, the install refuses an interpreter without reef, reef-pi doctor by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/389
refactor(evaluation): one plugin class per candidate-evaluation policy, composed by mixins by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/380
fix(records): retry the WAL switch that racing store openers lose by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/394
fix(docs): load site fonts with next/font by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/397
refactor(records): use SQLAlchemy Core and Alembic by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/395
perf(ci): parallelize tests and cache Python dependencies by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/398
docs(site): add an unlisted deck page at /deck by @quao627 in https://github.com/Human-Agent-Society/reef/pull/392
docs(site): serve the slides at /slides and redirect /deck by @quao627 in https://github.com/Human-Agent-Society/reef/pull/400
docs: the WeChat group is full, so the page shows the assistant's QR code by @quao627 in https://github.com/Human-Agent-Society/reef/pull/399
docs(site): slides number the loop steps and color the key terms by @quao627 in https://github.com/Human-Agent-Society/reef/pull/402
fix(harness): reef-pi doctor probes the release catalog, the route the API platform exposes by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/403
chore: point third_party/reef-client at the https upstream fix by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/406
feat(storage): decouple scenario stores and add PostgreSQL records by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/407
docs(profiles): say what selection always does in the harness-evolve profile by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/409
fix(harness): carry a scenario name with metacharacters verbatim into the wrapper by @wangzhengzhuo05 in https://github.com/Human-Agent-Society/reef/pull/382
feat(train): let a colocated LoRA step keep the frozen base resident by @HectorHHZ in https://github.com/Human-Agent-Society/reef/pull/393
test(native): wait for mount event after manifest retry by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/411
docs(site): slides draw their connector lines at an even pace by @quao627 in https://github.com/Human-Agent-Society/reef/pull/410
refactor: simplify scenario storage and remove package dependency cycles by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/412
fix(deploy): resolve local configs and report startup failures by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/413
refactor(config): unify service CLI and YAML parsing by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/415
refactor(config): share parsing across recipes and runtime adapters by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/416
docs: redraw the README architecture diagram as an animated SVG by @quao627 in https://github.com/Human-Agent-Society/reef/pull/418
docs: larger type in the README architecture diagram by @quao627 in https://github.com/Human-Agent-Society/reef/pull/419
feat(serve): unify config layout and launch inference without YAML by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/417
docs(coral): document how a training step works by @Xuan-1998 in https://github.com/Human-Agent-Society/reef/pull/421
refactor(coral): move recipe and demo under beta by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/423
feat(config): assemble components through backend definitions by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/424
feat(recipe): ship Reefine and document beta recipes by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/427
docs: one recipe catalog by task type, evolved object, and benchmark by @MikeZheng777 in https://github.com/Human-Agent-Society/reef/pull/429
docs: add emoji to the README navigation line and section headings by @quao627 in https://github.com/Human-Agent-Society/reef/pull/437
docs(site): redesign the home page around live repository activity by @quao627 in https://github.com/Human-Agent-Society/reef/pull/436
fix(deploy): reject non-finite ready timeouts by @YIKUAIBANZI in https://github.com/Human-Agent-Society/reef/pull/439
docs(site): plainer section titles on the home page by @quao627 in https://github.com/Human-Agent-Society/reef/pull/440
docs: add GEPA package README by @simonucl in https://github.com/Human-Agent-Society/reef/pull/433
docs(site): the home page ticker carries only news by @quao627 in https://github.com/Human-Agent-Society/reef/pull/441
feat(deploy): startup settings report, strict null and repeated YAML keys by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/442
examples: add runnable Meta-Harness Terminal-Bench campaign by @simonucl in https://github.com/Human-Agent-Society/reef/pull/434
refactor: simplify reported feedback and validate references at admission by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/445
refactor: unify training data and simplify report processing by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/446
chore(deps): bump mermaid from 11.17.2 to 12.0.0 in /docs/site by @dependabot[bot] in https://github.com/Human-Agent-Society/reef/pull/451
chore(deps-dev): bump ruff from 0.16.6 to 0.16.7 in the python-minor-and-patch group by @dependabot[bot] in https://github.com/Human-Agent-Society/reef/pull/449
fix(harness): put the harness binary's directory first on PATH for the session by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/453
chore(deps): bump the docs-site-minor-and-patch group across 1 directory with 7 updates by @dependabot[bot] in https://github.com/Human-Agent-Society/reef/pull/450
feat(harness): say when a release waits for a person's review by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/460
feat(reefine): a request gets a plan call, then the tool a step needs beside its rule by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/461
feat: add Tinker training backend and TTTD adapter by @jayzou3773 in https://github.com/Human-Agent-Society/reef/pull/448
refactor(tinker): fix dependency direction, one loss-family table, custom-loss hook by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/464
feat: let Reef own inference and training coordination by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/432
fix: propagate chat template errors and simplify runtime contracts by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/472
feat(sao): run CEO-Bench through Reef and train it with SAO in the episode by @MikeZheng777 in https://github.com/Human-Agent-Society/reef/pull/471
refactor(sao): rename the IMOAnswerBench example directory to imo_answerbench by @MikeZheng777 in https://github.com/Human-Agent-Society/reef/pull/473
docs: call README profiles deployment configurations by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/476
feat(train): write a task as a Harbor directory and split tasks by source by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/475
feat(observability): send operational metrics to W&B by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/480
feat(recipe): the harness gate takes its tasks from a split manifest by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/478
.github/workflows: Migrate workflows to Blacksmith except sandbox test matrix by @blacksmith-sh[bot] in https://github.com/Human-Agent-Society/reef/pull/490
docs: align README benchmark result links by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/489
perf(ci): shorten CI with separate sandbox suites and eight source workers by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/495
ci: separate routine PR checks from full validation by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/507
fix(core): preserve streamed messages in ATIF trajectories by @wbx398 in https://github.com/Human-Agent-Society/reef/pull/509
feat(train): define task generation processor contract by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/512
refactor(train): unify preparation under algorithm objectives by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/513
fix(tttd): preserve training timeout errors by @Chonghe-Jiang in https://github.com/Human-Agent-Society/reef/pull/384
feat: the task player plays Harbor tasks through Reef; SPADE writes Harbor tasks, runs one generation end to end and trains the Reasoning Agent on Tinker by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/494
feat(train): a shipped verifier reward scorer and the task on samples built from task reports by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/497
Reefine: close the loop with the person who asked by @quao627 in https://github.com/Human-Agent-Society/reef/pull/468
fix(record2dataset): the reply gate accepts dotfiles and refuses a COPY of a file the reply does not hold by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/517
feat(record2dataset): the Designer rules name the COPY context, the reward on every path and the package names by @Benjamin-eecs in https://github.com/Human-Agent-Society/reef/pull/519
Add team member Shiyi Zoe Du and Xinkai Zou by @quao627 in https://github.com/Human-Agent-Society/reef/pull/520
fix(harness): a reef-<adapter> run keeps its sessions so resume finds them by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/523
fix(harness): a reef-<adapter> run keeps the settings its first run saved by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/524
feat(pi): installing a release is the one decision, and a settled step offers it by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/525
docs: document inference timeout and retry deadline keys by @jane-jhu in https://github.com/Human-Agent-Society/reef/pull/526
fix(reefine): the built-in profile requires a token only when REEF_TOKEN is set by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/527
feat(inference): serve OpenAI Responses at /v1/responses by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/528
feat(reefine): clarify a request beside the session, answer it again when the review finds it short, and follow a Reef upgrade in reef's own entries by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/529
chore: ignore a harness tree the install script wrote by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/530
fix(slime): two breakages of the paper-scale SAO stack on the pinned slime by @Xuan-1998 in https://github.com/Human-Agent-Society/reef/pull/534
feat(connect): start Reef with --serve and report why Reef does not answer by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/535
fix(megatron): train without dropout unless the model or the CLI asks for it by @Xuan-1998 in https://github.com/Human-Agent-Society/reef/pull/536
feat(training): distillation training backend & pipeline by @MikeZheng777 in https://github.com/Human-Agent-Society/reef/pull/533
docs(readme): add the Trendshift badge by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/540
feat: multimodal provider calls as learning signal, and a coding-agent proposer that runs its change by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/538
perf(training): share prefix caches wherever they are cleared at publication by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/206
fix(harness): open the reef-pi request widget with ctrl+q, a key every terminal reports by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/542
feat(harness): /evolve and /versions with vN labels; tell the reefine agent when a decision model fits by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/543
fix(tutorial): repair Colab harness notebook startup by @JeyJeyGao in https://github.com/Human-Agent-Society/reef/pull/544
fix(harness): make version history easier to scan by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/546
fix(reefine): require native slash command integration in prompts by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/547
feat(harness): use evolve for the reef-pi command by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/549
feat(connector): support deleting connected scenarios by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/551
feat(sdft): the SDFT recipe with the skill stream as its example by @MikeZheng777 in https://github.com/Human-Agent-Society/reef/pull/539
feat(observability): export accepted records and commits as OpenTelemetry spans by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/548
fix(reefine): concurrent tool calls pull the agent's workspace one at a time by @quao627 in https://github.com/Human-Agent-Society/reef/pull/556
feat(reefine): ask only what a default cannot settle, mark the recommended option, and end the pages with the design and how to use the change by @quao627 in https://github.com/Human-Agent-Society/reef/pull/557
fix(reefine): refuse an extension that writes over the session's terminal, and say when a review did not run by @quao627 in https://github.com/Human-Agent-Society/reef/pull/560
chore(deps-dev): bump the python-minor-and-patch group with 2 updates by @dependabot[bot] in https://github.com/Human-Agent-Society/reef/pull/558
chore(deps): bump the docs-site-minor-and-patch group in /docs/site with 5 updates by @dependabot[bot] in https://github.com/Human-Agent-Society/reef/pull/559
feat(harness): requires kind binary, a program a release names by @Kaustubh1235 in https://github.com/Human-Agent-Society/reef/pull/553
feat(harness): require setup before starting agents by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/562
docs(harness): restructure the harness evolution loop section by @yanxz in https://github.com/Human-Agent-Society/reef/pull/561
perf(connector): report completed commands without heartbeat delay by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/564
fix(reefine): recover E2B disconnects and surface failed proposals by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/565
fix(harness): give a pi tree Reef's reply budget, and name a reply cut at it by @quao627 in https://github.com/Human-Agent-Society/reef/pull/566
fix(harness): keep an E2B sandbox alive past the ceiling E2B grants at creation by @quao627 in https://github.com/Human-Agent-Society/reef/pull/569
feat(reefine): add scripted trials and proposer progress by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/572
fix(reefine): make the proposer prove a change by its effect, not by the call returning by @quao627 in https://github.com/Human-Agent-Society/reef/pull/575
feat(reefine): ask which side runs when the machine and a paid model both serve a capability by @quao627 in https://github.com/Human-Agent-Society/reef/pull/577
fix(deploy): avoid Ctrl-C deadlock and allow forced shutdown by @BobbyZhouZijian in https://github.com/Human-Agent-Society/reef/pull/578
fix(record2dataset): validate reward map values in play_from_document by @KeWang0622 in https://github.com/Human-Agent-Society/reef/pull/574
[Refactor] Extract the chat facade from the SGLang handler by @ruizhang0101 in https://github.com/Human-Agent-Society/reef/pull/570
fix(harness): align the tree to the committed entries in CordisBackend by @simonucl in https://github.com/Human-Agent-Society/reef/pull/581