v1.3.0

v1.3.0

Everything merged since v1.2.0 (#136-#146). Full detail in CHANGELOG.md.

Skill #2 rewritten to Studio's real "new model" wizard

olmoearth-studio-job-config (vendored submodule, bumped to 2imi9/OlmoEarth-Skills@6839f1a) now walks the full wizard in order — Model name, Model type (with the fine-tuned-vs-embeddings framing), Foundation model, Label field, Training data (full vs a proportion), Data split (Spatial recommended), Temporal context, Image sources, Surrounding area, Review. The patch default is corrected 320 -> 640 m, and the skill states the core rule: Studio trains on Ai2's compute — it never asks about local hardware and never reroutes Studio model creation to local training (routing enforced in #142). SkillOpt-validated on the regenerated wizard-faithful benchmark: 0.714 hard / 0.906 soft (the previous skill scores 0.357 on the same targets).

New tools

• olmoearth_compare_group (skill #4): compare 2-6 prediction results pairwise + as an ensemble over their shared extent — agreement matrix, consensus fraction, most divergent pair, contested-point hotspots (#145).
• olmoearth_pixel_value (skill #4): read one result's model output at a point (#136).
• olmoearth_ensemble_uncertainty (skill #9): ensemble-disagreement confidence across distinct results (#136).
• QGIS .qlr layer definition + GDAL_WMS/XYZ descriptor + honest user-run COG recipe (skill #11, #136).

Security

• Path-traversal guard: model-controlled tool file I/O confined to a workspace root (security/paths.py, OLMOEARTH_OUTPUT_ROOT) (#137).
• Egress redirect re-validation on litsearch/HF fetchers; arXiv pinned to https (#137).
• Symmetric egress guards on the local and Anthropic LLM clients; QGIS tile-URL host gate (#136, #140).

Fixed

• "Configure a model" routes to the Studio wizard, not local rslearn/compute (#142).
• litsearch no longer overflows the local model's context on multi-call sessions (500-char abstracts, explicit search budget, one-16k-slot serving default) (#144).
• ok-envelope masked tool failures; docs aligned (#139, #140).

Web UI

• Results tab (#138); sidebar reorder + thin scrollbars (#141).

Docs

• docs/SHOWCASE.md regenerated live on this release tree — every transcript is a real agent-loop run (the skill #2 section embeds the new wizard SKILL.md; the #16 run exercises the new path-guard workspace root).

Verified on the release tree: ruff + mypy clean, 502 tests passing.

v1.2.0 - UI polish + skill slash-commands

Polishes the OlmoEarth Agent web UI and adds in-chat skill slash-commands.

New

• In-chat skill slash-commands — type / in the composer to call a specific skill directly (like Claude Code's /commands): a filterable menu of the 16 skills, arrow-nav + Enter/click to select. The brief is then routed to that skill (the displayed message stays as you typed it).

Fixed (from a multi-agent doc↔code↔architecture audit — every finding adversarially verified, 26 raw → 7 confirmed)

• README: "18-skill catalog" → 16; documented the 5 omitted js/ modules and the 7 omitted bridge endpoints.
• prefers-reduced-motion now covers the remaining animations (modal/detail entry, typing dots, caret, tool-call pulse, flash-glow).
• Added a --mint-hi design token; the difference-scan button hover uses it instead of a hardcoded color.

Also shipped in this line (the UI-polish pass)

Loading & busy states · difference-map radar scan · in-conversation result block · consecutive-workflow indicator · reproducible difference scan (retries flaky pixel fetches so the same pair gives the same numbers) · trajectory-chart fix · de-duplicated compare A/B legend · feature-showcase demo (webui/demo/record_showcase.py).

v1.1.0

Post-1.0 checkpoint: two new skills (catalog 15 -> 17), the make serve cache fix, and the sandbox / GDM / quick-start documentation pass.

Added

• Skill #16 olmoearth-litsearch: arXiv + OpenAlex literature search and DOI / arXiv-id resolution, so the agent can ground EO/geospatial citations in real papers instead of world-knowledge or hallucinated links. Key-free in-process tool bundle, deduped across sources (#62).
• Skill #17 olmoearth-automate: auto-decides embeddings vs fine-tuning for an EO task and proposes a config (model size, classifier head, embeddings-notebook command, fine-tune schedule, Studio job-config hand-off). Reuses the vendored olmoearth-embeddings decision table and can introspect a Hugging Face dataset (#58).
• docs/science-skills-assessment.md: a multi-agent assessment of Google DeepMind's science-skills, re-confirming the bundle is off-domain and that literature search is the one transferable capability (#69).

Changed

• Quick start reworked around the live web UI; the LLM is positioned as a "local or cloud API" choice (#66).
• README skill count corrected to 17 (#71).
• olmoearth-embeddings (#4) and olmoearth-automate (#17) disambiguated in README + SKILLS so the two Configure skills no longer read identically: #4 is the guidance + runnable-notebook generator, #17 is the one-call auto-decide + config that reuses #4's logic (#73).
• Corrected the system:python sandbox spec across docs (opt-in subprocess, not a preloaded persistent interpreter) (#68).

Fixed

• make serve reuses the cached model: the compose bind-mounts the host Hugging Face cache so a second make serve skips the ~17.7 GB re-pull (#67).

Full changelog: v1.0.0...v1.1.0

OlmoEarth Agent v1.0.0

OlmoEarth Agent v1.0.0

A natural-language agent that drives OlmoEarth Studio: 15 EO skills across prep, configure, run, analyze, integrate, and report; a local Qwen3.6-35B-A3B LLM by default plus optional bring-your-own-key Claude / ChatGPT / Gemini backends with model autodetect; a live project tree that labels models vs. embeddings; and a hardened FastAPI web bridge.

Licensed under the OlmoEarth Artifact License (Ai2). Hardened via a multi-agent readiness review (34 findings, 0 high-severity). See CHANGELOG.md.