[JSdotNet]

Summary

backlog-import-plan now favours prompt entries over work the user does by hand. It asks its questions while it writes the plan, so running a step later needs no manual step and no prompt stops to ask. This was a user request: "I prefer prompt type tasks over tasks that I have to do. Ask me questions during creation of the plan."

Changes

• SKILL.md (plugins/backlog-tools/skills/backlog-import-plan/)
  • New Prompts first section: prompt is the default kind. A task or test is only for the sign-off, or for a step the user chose to keep manual.
  • New Interview section:
    • Round one, before drafting: open decisions, values, access, and what a pass looks like. Each question offers a recommended answer first. It never asks for a secret's value, only where a session will find it.
    • Round two, after drafting: asks about each step still manual.
    • Further rounds whenever writing a prompt raises a question the answers don't settle.
  • Interview answers go into prompt instructions as settled facts, never "confirm with…", "decide whether…" or "TBD".
  • A manual test is drafted first as an AI-run QA prompt that starts the app, drives the scenarios and captures evidence.
  • Every kept task or test ends with a Kept manual: <reason> line in the user's words.
  • The closing report lists the answers, where each one went, and the steps kept manual.
• backlog-import-grammar.md: adds a paragraph saying prompts are the default and describing Kept manual: (body prose, not a token the app parses). The worked example's "confirm the format with design" decision task is replaced by a kept-manual task (reserving a marketplace publisher name).
• plan-review.html: two new checks.
  • Fails when a task or test other than the sign-off has no Kept manual: line.
  • Warns on open-decision wording in a prompt.
• README updated. backlog-tools version goes from 0.11.2 to 0.12.0.

Testing

• Built the review page from the updated worked example and opened it in a browser: all 38 checks pass.
• Built a second copy with the Kept manual: line removed and "TBD, confirm with design" added to a prompt: exactly one failure (the Kept-manual check) and one warning (open decision).
• No product code under src/ or tests/ changed, so there was no build or test run.

Related

No issue; this was an ad-hoc request. Builds on ADR 0007 (import reuses the entry-text grammar) and ADR 0013; neither changes.

Checklist

• Scope is focused and excludes unrelated changes
• Documentation or instruction files were updated when governance or behavior changed
• Running Aspire instances for this checkout were stopped before PR creation (none were started)
• Follow-up work is captured when the implementation is intentionally partial (none)
• Issues this PR closes are linked with a closing keyword, or there are none

🤖 Generated with Claude Code

[JSdotNet]

Merge review — ready

Severity	File	Finding	Action
Important	.claude-plugin/marketplace.json:11	backlog-tools is bumped to 0.12.0 in plugins/backlog-tools/.claude-plugin/plugin.json, but the marketplace entry still says 0.11.2. The two have moved together until now: 3921e82 changed both to 0.11.2 in one commit. The marketplace will advertise the old version.	Set the marketplace entry to 0.12.0 in this PR.
Important	plugins/backlog-tools/skills/backlog-import-plan/assets/backlog-import-grammar.md:75	The ## Entry kinds bullet still defines a task as "only the user does it: a decision, a sign-off, …". The paragraph this PR adds at line 89, and the new SKILL.md wording, say "A decision is never a task". The grammar now contradicts itself, and it is the file a model reads to learn the kinds.	Remove "a decision," from the task bullet, or point it at the new paragraph.
Suggestion	plugins/backlog-tools/skills/backlog-import-plan/assets/plan-review.html:488	The Kept manual: check accepts the line anywhere in the body. The spec says the body closes with it.	Test only the last non-blank instruction line, if placement matters.
Suggestion	plugins/backlog-tools/skills/backlog-import-plan/SKILL.md	The skill body grows from 204 to 244 lines, against the 40-line SKILL.md budget in the repository's conciseness rule. Most of the new lines are the Interview section.	Consider moving the Interview section to assets/ and linking to it.

Checks: all passing (CodeQL skipped) · Base: up to date (CLEAN) · Linked issue: none (ad-hoc request, stated in the body)
Conflicts with: none

Posted by schedule-merge-review on 2026-09-30T21:45Z. A person approves and merges.