AGENT SKILL

close-sprint

Validates and closes a completed sprint — verifies tickets, merges branch, archives sprint

License 许可
MIT
最近更新
2026-09-23
Stars
★ 0
Forks
0

01 前言 Front Matter

name
close-sprint
description
Validates and closes a completed sprint — verifies tickets, merges branch, archives sprint

03 SKILL.md 正文

Close Sprint Skill

This skill closes a completed sprint using the close_sprint MCP tool, which handles the full lifecycle.

Inputs

  • Active sprint directory in docs/clasi/sprints/NNN-slug/
  • All tickets for this sprint should be done

Process

  1. Gather sprint context: Call list_sprints() to identify the active sprint. Record the id and branch values — you will need them as sprint_id and branch_name in step 4. Do not proceed without these values in hand.

  2. Confirm with stakeholder: Present a summary of the sprint — list the completed tickets and key changes. Ask whether to proceed:

    • "Close sprint and merge to main" (recommended)
    • "Review completed work first"

    If the stakeholder chooses to review, invoke the sprint-review skill first.

  3. Load the tool schema: Call ToolSearch with query select:mcp__clasi__close_sprint to load the tool's parameter schema. This is required because CLASI MCP tools are deferred — calling them without first loading their schema causes all parameters to be silently dropped.

  4. Call close_sprint: Invoke the close_sprint MCP tool using the sprint_id and branch values collected in step 1:

    close_sprint(
        sprint_id="NNN",        ← from list_sprints() in step 1
        branch_name="sprint/NNN-slug",  ← from list_sprints() in step 1
        main_branch="master",
        push_tags=True,
        delete_branch=True,
        test_command="uv run pytest",  # or "" to skip tests
    )
    

    The test_command parameter controls how tests are run:

    • Omit or None: runs uv run pytest (default)
    • Custom string (e.g., "npm test"): runs that command
    • Empty string "": skips tests entirely (non-Python projects)

    The tool handles internally, in this exact order:

    • Pre-condition verification with self-repair
    • Run tests (if test_command is provided)
    • Archive sprint directory to sprints/done/
    • Update state DB, release execution lock
    • Apply the sprint's design overlay (opt-in only — see "Design Overlay Apply at Close" below)
    • Version bump and git tag
    • Merge to master, push tags, delete branch

Design Overlay Apply at Close

When Project.design_docs_opt_in is True and the sprint carries a design/ overlay (sprint.design_dir.exists()), close_sprint applies the overlay to the canonical docs/design/ doc set as one of its internal steps (design_overlay_apply), placed immediately after sprint.archive()/the DB update and before the version bump/tag step. Concretely it copies each overlay .md file (excluding .diff.md files) over its same-named canonical doc, then re-runs clasi design validate against the updated canonical set.

A failed apply or a failed post-apply validation blocks the version bump, tag, and merge — close_sprint returns an error result at the design_overlay_apply step and does not proceed further, the same fail-closed pattern already used for a failed test run. No partially applied doc set is ever tagged as a release. Recovery follows the same shape as any other blocked close: read the error, fix the underlying issue (a missing canonical target, a validation failure), and re-run close_sprint.

When opt-in is unset/False, or the sprint has no design/ directory (trivial/compact sprint under opt-in, or an opted-out project), this step no-ops silently — behavior is identical to today's close, no design/ overlay concept exists for that sprint.

  1. Report result: On success, report the version tag and merged branch. On error, report the blocker and recovery steps.

Issue Sweep at Close

When close_sprint runs, it automatically calls _sweep_done_issues, which moves any resolved sprint issues from <sprint>/issues/ to <sprint>/issues/done/. No manual move_issue_to_done call is needed for issues whose tickets are all done.

If any sprint issues remain unresolved at close, the close still succeeds. The result JSON will contain an unresolved_issues list with the filenames. Read this list and surface it to the team-lead for mop-up — these issues were not resolved in the sprint and need follow-up.

Issue Preconditions

Issues that are intentionally deferred (ticket carries completes_issue: false) pass cleanly through close without appearing in unresolved_issues. For all other in-progress issues, close collects their filenames in unresolved_issues and continues — the close is non-blocking on unresolved issues.

Resolution paths for issues that should have been resolved:

  • Tickets are done but issue not swept: check that all tickets referencing the issue carry issue: back-refs. If a back-ref is missing, call add_issue_ref and re-run close.
  • Issue has work remaining: call split_issue to split the remaining work into a new issue, then either defer it (it stays in the pool for the next sprint) or call create_ticket to bring it into the current sprint before closing.
  • Issue is intentionally deferred: set completes_issue: false on the ticket(s) referencing this issue. Close-sprint will then exclude that issue from unresolved_issues.

Output

  • Sprint branch merged to main and deleted
  • Sprint document moved to docs/clasi/sprints/done/
  • Sprint completion summary

04 同仓库更多 Skills