# Release Notes ## v0.2.2 - Local Evidence Workflow And Sanitized Regression Coverage Current stable release: v0.2.2. This release improves evidence quality without collecting customer data or changing the 12 published skill IDs. ### Changed - Expanded the deterministic real/sanitized regression library from seven to twelve cases, covering Power Query lineage, DAX rename impact, escaped MDX members, VBA control scope, and external dependency readiness. - Added an opt-in local JSONL observed-usage validator and aggregate-only summarizer. - Rejects paths, credentials, and customer artifact names in JSON keys and values; diagnostic paths redact sensitive key segments. - Added structural release-gate and capability-catalog coverage. The gate never requires a usage log or Excel. - Bumped the plugin manifest to `0.2.2+codex.20260714`. ### Boundary Observed usage is never collected, uploaded, or read from workbooks automatically. Synthetic regression cases validate deterministic contracts only; representative workbook-behavior proof remains a separate, local activity. ## v0.2.1 - Trigger Efficiency And Measured Discovery Cost Current stable release: v0.2.1. This release reduces plugin discovery cost and sharpens skill routing without changing the 12 published skill IDs or adding Excel functionality. ### Changed - Shortened the three manifest starter prompts to at most 110 characters. - Replaced all 12 canonical skill descriptions with compact `Use when ...` trigger conditions; skill bodies, routing behavior, IDs, and Excel feature scope are unchanged. - Added a 36-case customer-data-free trigger corpus with 24 positive and 12 confusable-negative cases, plus three real plugin-eval benchmark scenarios: `power-query-diagnosis`, `dax-versus-environment`, and `delivery-boundary`. Their inputs and generated responses are synthetic and do not prove real task success; observed usage is separate evidence. - Added the trigger validator to the structural release gate and capability catalog. - Synchronized generated `skills/`, `.claude/skills/`, and `.opencode/skills/` mirrors from `.agents/skills/`. - Bumped the plugin manifest to `0.2.1+codex.20260714`. ### Measured Comparison Static plugin-eval analysis of a newly staged runtime measured `trigger_cost_tokens` at 682 versus 1,161 for v0.2.0, a 41.26% decrease, while `invoke_cost_tokens` decreased from 15,365 to 14,886 (-479). These are synthetic/generated static estimates and do not prove real task success; observed usage is separate evidence. ### Reproduce The Static Comparison ```powershell $pluginEval = '' $runtime = Join-Path $env:TEMP 'excel-bi-v021-runtime' $before = Join-Path $env:TEMP 'excel-bi-v020-plugin-eval.json' $after = Join-Path $env:TEMP 'excel-bi-v021-plugin-eval.json' $compare = Join-Path $env:TEMP 'excel-bi-v020-v021-compare.md' python tools/build_runtime_package.py --project-root . --out-dir $runtime --require-pass node $pluginEval analyze $runtime --format json --output $after node $pluginEval compare $before $after --format markdown --output $compare ``` Keep the baseline, staged runtime, analysis JSON, comparison report, benchmark results, and observed-usage logs under the system temporary directory. See `docs/task-recipes.md` for the trigger validator and the three-scenario benchmark command. ### Validation ```bash python -m unittest discover -s tests -v python tools/validate-skills.py . python tools/run_release_gate.py --project-root . --profile structural ``` ### Boundary The trigger corpus and three benchmark scenarios use sanitized synthetic artifacts. Generated scenario responses do not prove real task success, and static token estimates are not observed usage. Real workbook success requires separately observed runtime and representative workbook-behavior evidence. ## v0.2.0 - Capability-Aware Excel Compatibility And Compact Runtime Release focus: make compatibility claims evidence-based across legacy, desktop, offline, Mac/web, Microsoft 365, and recipient environments while reducing the installed Codex runtime payload. ### Changed - Added a stable Windows capability probe contract covering Excel COM, workbook roundtrip, VBA project access, Power Query object-model/wait behavior, Data Model access, PDF export, ADO/ACE, MSOLAP, and ADOMD. - Added a cross-platform compatibility report that separates structural evidence, runtime capability evidence, and workbook behavior evidence; explicit capability requirements can block a handoff without treating an unavailable optional component as a package crash. - Added capability-first routing: platform/host/runtime availability requests go to `office-environment-diagnostics`, while DAX and Power Pivot formula/function compatibility remains with `power-pivot-dax-modeling`. - Added four synthetic capability fixtures and the `excel-capability-routing` sanitized regression case. No private workbook is required or shipped. - Added an allowlisted, deterministic runtime-package builder with SHA-256 inventory, dependency closure, compact README, deterministic zip, and private-artifact exclusions. - Added structural release checks for compatibility fixture/report behavior, runtime-package contents, release manifest version, and exactly three inspect/diagnose/publish starter prompts. The full Windows gate adds an optional live capability probe. - Documented Windows, macOS, Excel for web, Linux, Excel 2007/2010/2013/2016/2019, Office LTSC, Microsoft 365, 32-bit/64-bit, offline, WPS/LibreOffice structural-only boundaries, and authoring/automation/consumer/recipient targets. - Synchronized generated `skills/`, `.claude/skills/`, and `.opencode/skills/` mirrors from `.agents/skills/`. - Bumped the plugin manifest to `0.2.0+codex.20260714`. ### Validation ```bash python -m unittest discover -s tests -v python tools/validate-skills.py . python tools/run_release_gate.py --project-root . --profile structural ``` On Windows desktop Excel, the full gate also attempts the live capability probe: ```powershell python tools\run_release_gate.py --project-root . ``` ### Boundary Structural evidence does not prove Excel execution. Runtime capability evidence applies only to the probed machine. High-confidence compatibility requires representative workbook behavior evidence in each materially different target environment; third-party spreadsheet applications remain structural-only until tested in that host. ## v0.1.5 - GitHub Community Health And Safe Intake Release focus: reduce public collaboration risk by making issue intake, PR review, security reporting, and repository documentation surfaces safer. ### Changed - Added bilingual repository governance goals with objective, constraints, boundaries, can-do/cannot-do rules, detailed goals, and high-value backlog. - Added `CONTRIBUTING.md`, `SECURITY.md`, issue forms, and a pull request template. - Added `tools/validate_github_community_health.py` and included it in `node tools/install.mjs --check` and GitHub Actions. - Updated public checks across docs so community-health validation is part of the release gate. - Bumped the plugin manifest to `0.1.5+codex.20260623175347`. ### Validation Public checks: ```bash python tools/validate-skills.py . python tools/validate_project_docs.py --project-root . python tools/validate_github_community_health.py --project-root . python tools/validate_task_recipes.py --project-root . python tools/validate_official_docs_index.py --project-root . python tools/build_artifact_hygiene_report.py --project-root . --require-pass python tools/build_goal_coverage_report.py --project-root . --require-pass node tools/install.mjs --check ``` ### Boundary This release changes GitHub governance and public intake safety. It does not claim new Excel COM, VBA, Power Query refresh, or Power Pivot runtime proof. ## v0.1.4 - Public Growth Goals And Marketing Readiness Release focus: improve public trust, adoption clarity, social sharing, and marketing reuse without changing Excel workbook behavior. ### Changed - Added bilingual public growth goals with objective, constraints, boundaries, can-do/cannot-do rules, detailed goals, and high-value optimization backlog. - Added bilingual marketing copy pack for launch posts, short taglines, ad directions, channel variants, and do-not-claim rules. - Added maintainer attribution across README, project docs, Pages, and plugin manifest: Qiu Binbin (丘彬彬), WeChat `binstudy`, blog `90le.cn`. - Updated the website with use-case conversion cards, proof metrics, absolute Open Graph/Twitter image URLs, canonical URLs, author metadata, and v0.1.4 release visibility. - Updated project documentation validation and goal coverage checks so public growth, marketing copy, maintainer signature, and social metadata stay covered by CI. - Bumped the plugin manifest to `0.1.4+codex.20260623173419`. ### Validation Public checks: ```bash python tools/validate-skills.py . python tools/validate_project_docs.py --project-root . python tools/validate_github_community_health.py --project-root . python tools/validate_task_recipes.py --project-root . python tools/validate_official_docs_index.py --project-root . python tools/build_artifact_hygiene_report.py --project-root . --require-pass python tools/build_goal_coverage_report.py --project-root . --require-pass node tools/install.mjs --check ``` Maintainer structural gate: ```bash python tools/run_release_gate.py --project-root . --profile structural ``` ### Boundary This release changes public positioning, site layout, metadata, and documentation. It does not claim new Excel COM, VBA, Power Query refresh, or Power Pivot runtime proof. ## v0.1.3 - Public Maintenance Goals And CI Validation Release focus: reduce public repository maintenance risk without changing Excel workbook behavior. ### Changed - Added public maintenance goals, constraints, boundaries, can-do/cannot-do rules, detailed goals, risk register, and optimization backlog. - Added GitHub Actions public structural validation for pushes and pull requests. - Reworked project-document validation so it checks public repository docs instead of maintainer-only local ledgers. - Reworked goal coverage and completion readiness audits around public maintenance coverage and active backlog state. - Expanded `node tools/install.mjs --check` so it runs the full public structural check set. - Updated the website to expose release, public validation, and runtime boundary status. - Updated artifact hygiene expectations so the public package contains no Office workbooks by default. - Updated task recipes, distribution, open-source publishing, and real/sanitized regression docs to remove stale local-evidence assumptions. - Bumped the plugin manifest to `0.1.3+codex.20260623171436`. ### Validation Public checks: ```bash python tools/validate-skills.py . python tools/validate_project_docs.py --project-root . python tools/validate_github_community_health.py --project-root . python tools/validate_task_recipes.py --project-root . python tools/validate_official_docs_index.py --project-root . python tools/build_artifact_hygiene_report.py --project-root . --require-pass python tools/build_goal_coverage_report.py --project-root . --require-pass node tools/install.mjs --check ``` Maintainer structural gate: ```bash python tools/run_release_gate.py --project-root . --profile structural ``` ### Boundary This release does not claim new Excel COM, VBA, Power Query refresh, or Power Pivot runtime proof. Those checks still require Windows desktop Excel and task-specific runtime evidence outside the public repository. ## v0.1.2 - Split Bilingual Docs And Site - Split README and project overview into independent English and Chinese entry points. - Split the website into English and Chinese pages with browser-language redirect. - Kept install commands limited to supported Codex marketplace and local installer paths.