mirror of
https://github.com/github/spec-kit.git
synced 2026-07-04 04:45:43 +08:00
fd42fb15f4
* fix(taskstoissues): skip tasks that already have a GitHub issue Re-running /speckit-taskstoissues created a duplicate issue for every task because the command never checked for existing ones. Add a deduplication step before issue creation: list the repo's issues (state all) via the GitHub MCP server, collect the task IDs already present in issue titles, and skip any task that already has a matching issue. Issue titles are now prefixed with the task ID (e.g. T001:) so they can be matched on later runs, and list_issues is added to the command's MCP tools. Fixes #2968 * fix(taskstoissues): correct list_issues usage and issue title format Address Copilot review: - list_issues has no 'all' state; omitting state returns both open and closed issues. Use cursor-based pagination (after/endCursor) to fetch every page before building the dedup set. - task lines already start with their ID, so reuse the task text as the issue title instead of prefixing the ID again (which produced 'T001: T001 ...'). * fix(taskstoissues): match task IDs anywhere in titles and define one canonical title Address follow-up Copilot review: - task lines start with a markdown checkbox (- [ ] T001 ...), so the creation step now strips the checkbox and [P]/[US#] markers and writes a single canonical title 'T001: <description>'. - dedup now scans each issue title for a T<digits> token anywhere in the title, so existing issues titled 'T001 ...', 'T001: ...' or '[T001] ...' are all matched. * fix(taskstoissues): use word-boundary task ID match and request perPage 100 Address Copilot review: - match issue titles against \bT\d{3}\b so tokens like ST001 or T0010 are not matched by mistake (task IDs are T + 3 digits). - request perPage: 100 on list_issues to reduce pagination calls. * fix(taskstoissues): bound issue pagination to the tasks being processed Address Copilot review: extract the task IDs from tasks.md first, then paginate list_issues only until every task ID has been matched (or pages run out), instead of fetching the repo's entire issue history. Keeps the call count bounded on repos with large issue backlogs.