AI Sprint Review

Sprint Review

Purpose

Generate branded sprint review PowerPoint presentations using python-pptx. Each invocation produces a NEW python-pptx script tailored to the current sprint's data -- the script is regenerated every time, not reused from a static template.

Trigger

• Command: /ai-sprint-review
• Context: end of sprint, presenting delivered work to stakeholders.

When to Use

• End of sprint: presenting delivered work to engineers, heads, and managers
• Monthly reviews with stakeholders
• Ad-hoc progress reports for a specific period

When NOT to Use

• Daily updates -- use /ai-standup
• Sprint planning -- use /ai-sprint plan
• Retrospectives -- use /ai-sprint retro
• Incident write-ups -- use /ai-postmortem

Pre-conditions (MANDATORY)

Before gathering data, read configuration:

1. Read .ai-engineering/manifest.yml -- focus on the work_items section.
2. Determine active provider: work_items.provider is either github or azure_devops.
3. Read provider-specific config:
   • GitHub: work_items.github.team_label
   • Azure DevOps: work_items.azure_devops.area_path
4. Read quality thresholds from quality section (coverage, duplication, cyclomatic, cognitive).

Procedure

Step 1 -- Determine Sprint Period

Resolve the date range for data gathering:

Step 2 -- Gather Data

Collect data from three sources in parallel:

2a. Work Items (from provider)

GitHub:

gh issue list --label "<team_label>" --state all \
  --json number,title,state,labels,milestone,closedAt,assignees

Azure DevOps:

az boards query --wiql "SELECT [System.Id], [System.Title], [System.State], [System.AssignedTo], [System.WorkItemType] FROM WorkItems WHERE [System.AreaPath] UNDER '<area_path>' AND [System.IterationPath] = @CurrentIteration" --expand relations

Categorize each item:

• Completed: closed/resolved during the sprint period
• In Progress: open, with activity during the sprint
• Carried Over: open, existed before sprint start, no completion

Walk the hierarchy when available: Feature > User Story > Task.

2b. Git Activity

# Commits
git log --since="<start>" --until="<end>" --format="%h|%s|%an" --no-merges

# Merged PRs (GitHub)
gh pr list --state merged --json number,title,mergedAt,author

# Merged PRs (Azure DevOps)
az repos pr list --status completed

Derive: commit count, unique authors, LOC delta, files changed.

2c. Quality Metrics

# Test count
pytest --co -q 2>/dev/null | tail -1

# Lint status
ruff check . --statistics 2>/dev/null

Compare actuals against thresholds from .ai-engineering/manifest.yml quality section.

Step 3 -- Generate python-pptx Script

Generate a NEW script each time. Reference docs/presentations/generate_sprint_review.py for style conventions. Do NOT copy the old script verbatim -- adapt all content to the current sprint's data.

Brand Constants (MUST use these exact values)

AI_BG_DARK = RGBColor(0x0B, 0x11, 0x20)
AI_WHITE = RGBColor(0xFF, 0xFF, 0xFF)
AI_TEXT_PRIMARY = RGBColor(0xE2, 0xE8, 0xF0)
AI_ACCENT = RGBColor(0x00, 0xD4, 0xAA)
AI_PRIMARY = RGBColor(0x1E, 0x3A, 0x5F)
AI_PRIMARY_LIGHT = RGBColor(0x2A, 0x4F, 0x7A)
AI_ERROR = RGBColor(0xEF, 0x44, 0x44)
AI_SUCCESS = RGBColor(0x10, 0xB9, 0x81)
AI_WARNING = RGBColor(0xF5, 0x9E, 0x0B)
AI_TEXT_LIGHT = RGBColor(0xF8, 0xFA, 0xFB)
AI_NEUTRAL = RGBColor(0x64, 0x74, 0x8B)
AI_TEXT_MUTED = RGBColor(0x94, 0xA3, 0xB8)
AI_BORDER_DARK = RGBColor(0x1A, 0x2A, 0x40)
AI_CARD_DARK = RGBColor(0x1E, 0x29, 0x3B)
SEC_BLUE = RGBColor(0x2E, 0x6B, 0xA4)
SEC_PURPLE = RGBColor(0x7B, 0x3F, 0xA0)

Typography

• Titles/headings: JetBrains Mono
• Body text: Inter

Layout

• 16:9 aspect ratio: 13.333" x 7.5"
• LEFT_MARGIN: 1.2"

Helper Functions (include in generated script)

Include these helpers matching the signature patterns in the reference script:

• add_textbox -- single-style text box with font, alignment, anchor controls
• add_rich_textbox -- multi-line text box with per-line styling (font, size, color, bold)
• add_accent_bar -- thin colored horizontal bar
• add_card -- filled rectangle with optional border and left accent stripe
• add_rounded_rect -- rounded rectangle for badges
• _add_kpi_card -- card with large value + small label, centered
• add_slide_header -- accent bar + title + optional subtitle
• _blank_slide -- blank layout with dark background fill
• set_notes -- attach speaker notes to a slide
• add_styled_table -- table with colored header row
• _style_cell -- style a single table cell (text, font, alignment, fill)
• _add_feature_card -- card with title + bullet list and left accent

Slide Structure (8-14 slides)

Content Rules

• Technical but accessible -- audience is engineers, heads, and managers
• Include speaker notes for EVERY slide (the set_notes call)
• Quantify everything: numbers, percentages, before/after comparisons
• Use accent colors to differentiate categories (AI_ACCENT for primary features, SEC_BLUE/SEC_PURPLE/AI_WARNING/AI_ERROR for variety)
• Feature deep-dive slides use _add_feature_card for consistent card layout
• KPI cards use _add_kpi_card for large-value display

Step 4 -- Execute and Output

1. Write the generated script to docs/presentations/generate_sprint_review.py (overwrite the existing file).
2. Run: python3 docs/presentations/generate_sprint_review.py
3. Output file: docs/presentations/sprint-review-YYYY-MM.pptx
4. Report to user: slide count, file path, key stats (commits, PRs, coverage).

Arguments

Quick Reference

/ai-sprint-review                          # current month, auto-detect provider
/ai-sprint-review --sprint 2026-03         # March 2026 sprint
/ai-sprint-review --iteration "Sprint 12"  # named iteration from provider

Common Mistakes

1. Reusing the old script -- every invocation MUST generate a new script with current data. Never copy-paste the previous script unchanged.
2. Missing speaker notes -- every slide must have set_notes(). Stakeholders use presenter view.
3. Wrong color palette -- use the brand constants above, not arbitrary colors.
4. Skipping pre-conditions -- always read manifest.yml first to get provider config and quality thresholds.
5. Hardcoding dates -- derive the sprint period from arguments or current date, never hardcode.
6. Missing python-pptx import -- the generated script must import from pptx and its submodules.

Integration

• Reads work_items config from .ai-engineering/manifest.yml
• Can reference active specs from .ai-engineering/specs/
• Quality thresholds from manifest.yml quality section drive the metrics table
• Git history and PR data provide quantitative backing for every claim

$ARGUMENTS