ORE Studio Sprint 20 – Release Notes

• 31 workflow skills added: every compass pillar (bearings, sprint, PR, review, task, story, timeline, capture, build) now has a dedicated action-oriented skill; the naming convention and skill map were also established, disambiguating skills from recipes and runbooks.
• Create vs pick-up split: scaffolding a story/task is now separate from clocking on to it; bookkeeping (state flip, result, journal) always rides the PR rather than needing its own follow-up.
• compass quality of life (15 improvements): SSH_AUTH_SOCK from .env, dashboard services bug fix, Claude settings auto-authorisation, actionable search results, journal log newest-first, capture --commit, folder-name search, task move, recipe and inbound-link ranking boosts, PR review gate, cross-reference wiring, story tasks subcommand, scoped search (--under/--type), capture promote fix.
• compass heading: new command ranks candidate next work items from journal activity, fleet state, sprint state, and backlog buckets; optional keyword steering.
• compass site serve: serve-site.sh deleted; the Python HTTP server now lives in compass site serve with --compile and --port options and dashboard integration.
• compass test run: new subcommand runs all tests, composite or per-component targets, or Catch2 filters; --build delegates to cmake; .env loaded for direct binary runs.
• Sprint timeline: compass timeline snapshot generates bucketed activity org files from git/PR/agile history; timeline catalogue indexes them; the agile board gained a timeline view tab.

• All 32+ mustache templates in literate org: each template lives in a facet-documenting org doc with the mustache source as a tangled block; a three-level hierarchy links template → facet → group → overview; zero-diff drift verified at migration.
• MASD knowledge graph complete: MASD methodology and applied docs restructured; terminology (projection, facet, facet group) corrected throughout the literate template library; doc facet catalogue cross-linked to template org files and into org-roam.
• MASD model catalogues: module catalogues for all floating entity models; component overview pages catalogue modules and sub-components; variability model rewritten to separate activation from configuration.
• Doc format cleanup: #+version field stripped from all org files (800+ files); doc_version_audit.py retired; stale v2 codegen terminology purged.
• Knowledge graph filter: org-roam-ui moved to external/ (no patches needed); :skill: and :llm: filetags applied to all LLM docs; tag export fixed to use ALLTAGS; LLM view filter works from the graph UI.

• Design: the three Qt provisioner wizards (system, tenant, party) mapped to idiomatic ores-shell commands with a comprehensive interface brainstorm and coverage analysis.
• Implementation: flag parsing, stop-on-error load, full plumbing (bundles, workflow, lei, synthetic, parties, account-parties, reports), and top-level provision system/tenant/party porcelain.
• Script library: org-mode sources with detailed commentary in projects/ores.shell/scripts/; emacs-tangle CMake target; provision_all end-to-end script; user manual chapter added.

• Built-in help system: user manual HTML exported via org-publish with Qt-help-friendly configuration; .qhp project and .qch collection built by qhelpgenerator; HelpViewer widget backed by QHelpEngine wired into the Help menu. Offline, searchable, and always in sync with the published manual.

• Structured site navbar: flat twelve-item nav replaced with a hierarchical dropdown menu; Agile merged into Developer; items reordered for new users.
• Edit on GitHub button: every published page carries a link to the source org file in the GitHub web editor — no more copy-pasting paths.
• Site preview port: ORES_SITE_PORT moved from ad-hoc 8000+suffix into the environment's base port range (+4), matching all other env-specific ports.

• CI retuned: pr-gate single required check with path-classified jobs (C++, shell, site, agile); main matrices staggered at 2h/3h/4h with a doc-only skip guard; Gemini review replaced by Claude via a new claude-review.yml workflow; branch protection pointed at pr-gate.
• Skip vcpkg for doc builds: ORES_VCPKG_ENABLED cmake cache variable gates the entire vcpkg/C++ toolchain; linux-doc-only preset flips it off; site CI no longer restores a 2 GB sccache it never uses.
• CDash visibility: PR notes field links canary entries to their PR; site builds appear in a dedicated Site category; Experimental dropped from developer-facing CI.
• Fix flaky IAM tests: UUID v7 timestamp collision in email generation caused duplicate-key failures under parallel CI; one-line fix in account_generator.cpp.
• Hotfix — CI disk: ubuntu-24.04 runners ran out of disk during canary builds; expanded cleanup step removes linuxbrew, ghcup, miniconda, Azure/GCloud CLIs, PowerShell, Chromium, snap/lxd, node_modules, and sccache from the site workflow.

• JavaScript agile board: ores.org-js single-page dashboard deployed with the site; parses graphdata.json from the org-roam export; renders sprints, stories, and tasks with a timeline tab.
• Memory migration: Claude harness memory files audited and ported to the org-based memory system; auto-memory disabled in Claude Code settings; harness files deleted.
• Sprint audit and health charts: audit ran clean at every health review and at close; four chart types regenerated at sprint close with gnuplot.

Dual-axis bar chart. PRs (left axis) and commits (right axis) per day. A high commits-to-PR ratio may indicate scope creep.

Six-day sprint with a clear build-up: days 1–2 (06-06 to 06-07) were provisioning and skill work landing in bursts; the peak is 06-10 to 06-11 when the compass QoL round, heading command, timeline, and CDash stories all merged in rapid succession. The commits-to-PR ratio stays healthy throughout — small, well-scoped PRs rather than large squashes.

Lines added (green) and deleted (red) per day.

Additions dominate on most days, matching the provisioning and compass feature work. The largest deletion day is 06-08 to 06-09 — the format-cleanup story stripping #+version from 800+ files — giving the characteristic refactoring signature of high deletions with moderate additions. No single day overwhelms the chart the way sprint 19's mass-format pass did.

Hours from PR open to merge.

The majority of PRs merge in 2–6 hours. Outliers correspond to PRs that waited on a review round from the Claude bot — the review gate introduced mid-sprint adds 1–2 hours to the cycle but catches real issues. A handful of late-sprint PRs (hotfixes, close-out) merge within an hour. No PRs stalled for more than a working day.

Stories marked DONE during the sprint.

Steady upward slope across all six days — roughly 5–6 stories per day — reflecting a sprint where most stories were genuinely small and self-contained. The slope does not accelerate at the end the way sprint 19's did (where late health reviews closed a backlog of already-merged work). Stories were closed promptly as their PRs merged.