Sprint Review
Evaluate a sprint's readiness before starting it. Detect quality issues across all tickets and produce a structured report with actionable recommendations.
Language
Instructions in English (for LLM processing). ALL user-facing output MUST be in Spanish, following the format in references/output-format.md.
Configuration
- Cloud ID:
afianza-ac.atlassian.net - Default project:
DEVPT - Oversized threshold: 5 story points
- Systemic threshold: 50% of filtered tickets (findings above this become systemic)
- Vague terms (ES): "debería", "quizás", "posiblemente", "varios", "algunos", "etc.", "por definir", "TBD", "en general", "eventualmente", "se podría", "a futuro"
- Vague terms (EN): "should", "maybe", "possibly", "various", "some", "etc.", "TBD", "in general", "eventually", "could be", "as needed", "appropriate"
Instructions
Phase 1: Get sprint tickets
Parse $ARGUMENTS to determine input mode:
Mode A — Sprint from Jira (default):
If no arguments or a board ID is provided, fetch tickets from the next upcoming or active sprint.
ToolSearch → "select:mcp__claude_ai_Atlassian__searchJiraIssuesUsingJql"
JQL options (try in order until one returns results):
project = DEVPT AND sprint in futureSprints() ORDER BY rank ASC(next sprint)project = DEVPT AND sprint in openSprints() ORDER BY rank ASC(active sprint)
If a board ID is given as argument, use: sprint in futureSprints() AND project = DEVPT
Mode B — Manual ticket list:
If arguments contain comma-separated ticket keys (e.g., DEVPT-100,DEVPT-101,DEVPT-102), use those directly.
ToolSearch → "select:mcp__claude_ai_Atlassian__getJiraIssue"
Fetch each ticket individually using mcp__claude_ai_Atlassian__getJiraIssue(cloudId: "afianza-ac.atlassian.net", issueIdOrKey: "<key>").
For both modes, extract from each ticket:
- Key, summary, description, acceptance criteria
- Issue type (Story, Bug, Task, Tech Debt, Sub-task)
- Story points (use
customfield_10031— NOTcustomfield_10016which is always null) - Status, assignee, labels
- Issue links (blocks, is blocked by, duplicates, relates to)
- Sprint name (for report header)
If no tickets are found, tell the user and stop.
Phase 1.5: Filter tickets
Before running checks, filter the ticket list:
- Exclude completed tickets — Remove any ticket with status
Done,Closed, orDesestimada. These are not actionable. - Exclude sub-tasks — Remove issue type
Sub-task. They are covered by their parent ticket. - Record the original total count for the header (e.g., "50 tickets total") but only analyze the filtered set (e.g., "24 tickets analizados").
Phase 2: Run checks
Execute ALL checks against the filtered ticket list. Each check produces findings with a severity level.
Check 1: Duplicate detection
Detect tickets that describe the same deliverable or scope of work — i.e., doing one would make the other redundant.
How to evaluate:
- Read the title AND description of each ticket to understand what it delivers.
- Compare pairs only when titles suggest they could be about the same thing.
- Flag as duplicado potencial ONLY when you can explain concretely why both tickets would produce the same outcome or change.
What is NOT a duplicate:
- Tickets in the same functional area but with different deliverables (e.g., "Importar datos empleados baja" vs "Incluir campos apartado baja empleado" — one imports data, the other adds UI fields).
- Tickets assigned to the same person in a similar status — that alone means nothing.
- Tickets that share keywords but do different things (e.g., three document tasks that each fix a different aspect).
When there is not enough information: If two tickets have similar titles but lack description/AC to determine whether they overlap, do NOT flag them as duplicates. Instead, note them in a separate "Requieren mejor descripción para diferenciar" list — the problem is poor definition, not duplication.
Severity: Critical (only for confirmed potential duplicates) Output per finding: The ticket keys, their titles, and a concrete explanation of WHY they appear to deliver the same thing.
Check 2: Definition quality
For each ticket, check:
| Validation | Fails when | Severity |
|---|---|---|
| Description missing | No description or < 20 words | Critical |
| Acceptance criteria missing | No ACs found (no "criterios", "AC", "acceptance", checklist, or Gherkin) | Critical |
| Description is vague | Description exists but is mostly high-level without concrete details (< 50 words of substance) | Warning |
Output per finding: Ticket key + which validation failed.
Check 3: Estimation gaps
For each ticket, check if story points are assigned. All tickets must have story points before entering a sprint.
Story-level rule: If a Story has sub-tasks, its story points MUST equal the sum of its sub-tasks' story points. Flag mismatches as Warning with the expected vs actual values.
Severity: Critical (missing SP), Warning (sum mismatch) Output per finding: List of unestimated ticket keys. For mismatches: ticket key, current SP, sum of sub-tasks SP.
Check 4: Vague language
Scan title + description + ACs for terms from the vague terms lists (see Configuration).
Severity: Warning Output per finding: Ticket key, the vague term found, and the surrounding sentence.
Check 5: External dependencies
For each ticket, check issue links:
- Links of type "is blocked by" pointing to tickets NOT in this sprint
- Links of type "is blocked by" pointing to tickets with status != Done/Closed
Severity: Critical Output per finding: Ticket key, blocking ticket key, blocker's status, whether it's in the sprint.
Check 6: Sprint balance
Calculate (using the filtered set, excluding sub-tasks and Done):
- Count and percentage of each issue type (Story, Bug, Task/Tech Debt)
- Total story points
- Average story points per ticket (excluding unestimated)
No specific severity — this is informational. Flag if:
- Bugs > 40% of sprint (Warning: "Sprint dominated by bug fixes")
- No stories at all (Warning: "Sprint has no new feature work")
- Total points significantly different from typical (if detectable)
Check 7: Oversized tickets
Flag tickets with story points > threshold (see Configuration).
Severity: Warning Output per finding: Ticket key, points, recommendation to decompose.
Check 8: Unassigned tickets
Flag tickets where assignee is null or empty.
Severity: Critical Output per finding: Ticket key. No ticket should enter a sprint without an owner.
Phase 2.5: Classify systemic findings
After running all checks, determine which findings are systemic:
- A finding is systemic if it affects >50% of the filtered tickets (see Configuration: systemic threshold).
- Example: if 30/35 tickets have no description, "sin descripcion" is systemic.
- Systemic findings are shown in the SCORECARD as
SISTEMICO: [N]/[T] sin descripcionand are NOT repeated per ticket in the table. - A ticket that has ONLY systemic findings does NOT appear in the table — it goes to "Sin hallazgos adicionales".
- A ticket appears in the table only if it has at least one differential (non-systemic) finding.
Phase 3: Build report
Organize findings using the format in references/output-format.md.
Key principle: ticket-centric. Each ticket appears ONCE in the table with ALL its differential findings aggregated. No ticket key should be repeated across multiple sections.
Order:
- Header (sprint name, tickets analyzed, total points)
- Scorecard (three lines: SISTEMICO counts, CRITICO counts, WARNING counts)
- "Requieren mejor descripción para diferenciar" section (if any)
- Duplicates section (pairs/groups with explicit motivo)
- Tickets table (bordered, 5 columns, sorted by severity — only differential findings)
- Sin hallazgos adicionales (tickets with only systemic findings)
- Resumen de recomendaciones (numbered, prioritized actions)
Phase 4: Offer actions
After displaying the terminal report, ask the user which actions to take (multiple choice):
- Enviar a Slack — Send the report as a Slack draft message using the format in references/output-format-slack.md. Ask the user for the channel name, then use
mcp__claude_ai_Slack__slack_search_channelsto find the channel ID, andmcp__claude_ai_Slack__slack_send_message_draftto create the draft. If the report exceeds 5000 chars, split into multiple messages. - Comentar en tickets individuales — Add a comment on each ticket that has Critical findings, noting what needs attention before sprint start. Use
mcp__claude_ai_Atlassian__addCommentToJiraIssuewith cloudIdafianza-ac.atlassian.net. Keep comments concise and actionable. - Solo el reporte — No external actions, just the terminal output.
Edge cases
- If a ticket key is invalid, skip it and note it in the report.
- If JQL returns no results for both future and open sprints, ask the user for ticket keys manually.
- Do NOT invent findings. If a sprint is well-prepared, say so explicitly.
- Duplicate detection is heuristic — always phrase as "potential duplicate" and let the team decide.
- If story points field name differs, try common Jira field names before giving up.