BLEND Kit
BLEND Kit packages six reusable workflows for BLEND delivery: task generation, test specification, automation testing, artifact review, implementation planning, and code review.
The private repository is a marketplace named tryhand-blend-kit. The install ID is blend-kit@tryhand-blend-kit.
Install
Repository access is required because the marketplace is private.
Codex
codex plugin marketplace add git@github.com:TryHand-Co-Ltd/blend-kit.git
codex plugin add blend-kit@tryhand-blend-kit
codex plugin list --json
Restart Codex or open a new session after installation.
Claude Code
claude plugin marketplace add git@github.com:TryHand-Co-Ltd/blend-kit.git
claude plugin install blend-kit@tryhand-blend-kit
claude plugin list
Restart the Claude Code session after installation. Skills are invoked with the plugin prefix, for example /blend-kit:blend-review-code.
Cursor
Open Customize → Plugins → From GitHub Repository, select TryHand-Co-Ltd/blend-kit, then install BLEND Kit. Teams and Enterprise organizations can add the same private repository as a Team Marketplace.
The repository includes .cursor-plugin/marketplace.json and a portable Agent Plugin manifest, so Cursor loads the same six skills from the generated package.
Included skills
| Skill | Purpose |
|---|---|
blend-generate-task | Generate BLEND task breakdowns and acceptance artifacts. |
blend-generate-test-spec | Generate or refresh Test Specs and editable JA/VI reports. |
blend-automation-test | Execute provided Test Cases with the available Playwright MCP, review appropriately scoped annotated image evidence, and update one XLSX or Google Sheets report. |
blend-review-artifacts | Review task and Test Spec artifacts against source requirements. |
blend-plan-implementation | Create an implementation plan from approved scope. |
blend-review-code | Review scoped code changes against requirements and project rules. |
Update
Codex
codex plugin marketplace upgrade tryhand-blend-kit
codex plugin remove blend-kit@tryhand-blend-kit
codex plugin add blend-kit@tryhand-blend-kit
For Claude Code or Cursor, refresh the marketplace and reinstall/update the plugin from that client, then start a new session.
Build and validate
pwsh -File scripts/build-plugin-package.ps1 -Python python
pwsh -File scripts/check-plugin-package.ps1
python tests/test_kit.py --area all
The generated package is stored at dist/plugins/blend-kit.
For a customer-facing report, use Generate Test Spec's default report (test-report@2.4.0). The same workbook supports testing and customer review: self-contained conditions/actions/branch-specific outcomes and independent result rows. Human descriptions precede variant codes. Only images expand, in one initially collapsed group per TC; descriptive titles sit above vertically stacked images, without evidence URL inputs. No default technical appendix, preparation label or review notes; internal readiness/oracle/provenance safeguards remain. Older workbook formats are rejected without rewriting; no migration helper is shipped. XLSX-to-Google-Sheets conversion after testing needs native readback and viewing checks before sharing.
Current authored source uses test-cases@1.3.0 and the only report layout is test-report@2.4.0, from the root JA/VI asset pair. Supported source1.0..1.2 remains readable with original IDs, literals, oracles and gaps; this does not retain old workbook templates or dispatch. Actual/Status stay with their TC, and unused image reserve rows are fully hidden. Browser evidence defaults to 1920×1080 and screenshots default to full_page=false, with full page only when needed. Short English observed notes sit top-right without covering UI; minimal highlights do not overlap/touch, and actual pixels are reviewed.
The automation skill selects one authoritative report per run and archives files by verified feature and Run ID under blend-context/local/test-output/. It does not upgrade Playwright MCP, redesign TC during execution, seed databases, modify the application, migrate existing reports, install skills or publish as part of a test run. Native Sheets needs actual connector/UI capabilities and separate readback/image verification, including grouping and image layout; XLSX/package checks do not certify those behaviors. The RC-001 report is a blank handoff for a separately authorized automation run; cleanup does not execute tests.
The previously verified package snapshot (before this source change):
- Plugin:
blend-kit@tryhand-blend-kit - Version:
1.0.0 - Skills:
5 - Files:
207 - Tree digest:
84f6d975077b7654ac4e77d0490644bc068be97a1270d470f66c0cd100c9eeae
See workflow rules, artifact formats, and package maintenance for implementation details.