Backlog grooming assistant

Backlog Grooming Assistant

This skill reads a local backlog export (CSV or Markdown table), flags problematic items using a structured checklist, and produces a grooming session agenda with a scorecard. No external service connections required — works entirely from pasted or uploaded file contents.

Input:

• Backlog file contents: CSV (with column headers) or Markdown table — pasted directly or uploaded
• Minimum required columns: item ID and title; additional columns (owner, priority, estimate, status, last-updated) enable more flags

Output:

• Markdown document with: Scorecard → Top Issues → Grooming Agenda → Skipped Checks

---

Language Detection

Detect the user's language from their message:

• If Russian (or contains Cyrillic): respond in Russian
• If English (or other Latin-script language): respond in English
• If ambiguous: respond in the language of the trigger phrase used

The detected language applies to all output text: section headings, table column headers, field labels, agenda topic titles, error messages, and recommended actions. Translate every part of the output format template into the detected language — do not leave any section header or label in English when responding in Russian.

---

Instructions

Step 1: Parse and Validate Input

1. Accept backlog content from the user (pasted text or file contents)

   • If content is empty: stop and return — "Backlog is empty — no items to analyze. Paste CSV or Markdown table content."

2. Detect format:

   • CSV: first line contains comma-separated headers
   • Markdown table: lines starting with | and separator row |---|
   • If neither format detected: stop — "Format not recognized. Supported: CSV with headers, Markdown table. Paste the file contents in chat."

3. Parse rows: extract all items as a list of key-value pairs

   • Identify available columns by name (case-insensitive matching)
   • Map common synonyms: "Task" → title, "Assignee" → owner, "SP"/"Points" → estimate, "Updated"/"Modified" → last-updated
   • If columns cannot be mapped automatically: ask user to specify column mapping for the ambiguous columns, then continue

4. Validate row count:

   • If 0 data rows (headers only): stop — "Backlog is empty — no items to analyze."
   • If all rows have status "Done" or "Closed": stop — "All items are completed — no active tasks for grooming."

Step 2: Apply Flag Checklist

For each item, evaluate applicable flags (skip flags that require unavailable columns):

Flag	Condition	Required column(s)
NO_OWNER	Owner/Assignee field is empty or missing	owner
NO_ESTIMATE	Estimate/story points field is empty or zero	estimate
UNCLEAR_SCOPE	Description/title is absent or fewer than 5 words	title, description
STALE	Last-updated date is more than 14 days ago	last-updated
BLOCKED_NO_ETA	Status contains "blocked" with no ETA/date mentioned	status, description
ALL_HIGH_PRIORITY	Signal: more than 50% of all items marked High/Critical	priority (applied to entire backlog, not per item)
DUPLICATE_TITLE	Title is ≥80% similar to another item in the backlog	title

Compile per-item flag list and a backlog-wide summary count for each flag type.

Edge Cases:

• Column missing for a flag: skip that flag for all items; record the skipped check in output
• STALE check: if date format is ambiguous, skip and note in skipped checks
• ALL_HIGH_PRIORITY: apply once to the whole backlog; annotate in scorecard, not per-item

Step 3: Identify Top Issues

1. Sort flagged items by flag count (descending)
2. Select top 10 items (or fewer if backlog is small)
3. For each top item: collect ID, title, flags, and priority level

Step 4: Generate Grooming Agenda

Group top issues by flag type into agenda topics:

• No Owner → agenda item: "Assign owners to unowned tasks"
• No Estimate → "Estimate unscored items"
• Unclear Scope → "Clarify scope / rewrite descriptions"
• Stale Items → "Review or close items with no recent activity"
• Blocked Items → "Resolve blockers or set ETA"
• Priority Inflation → "Re-prioritize — too many High/Critical items"
• Possible Duplicates → "Review and merge duplicate candidates"

For each agenda topic, list the affected item IDs and a recommended action.

Step 5: Produce Output

Write the full output in the format specified in the Output Format section below.

---

Negative Cases

• Empty input or headers-only file: Stop. Return "Backlog is empty — no items to analyze."
• Unrecognized format: Stop. Return "Format not recognized. Supported: CSV with headers, Markdown table."
• All items Done/Closed: Stop. Return "All items are completed — no active tasks for grooming."
• Plain text list without structure: Warn — "Insufficient structure for flag analysis. Try adding columns: status, owner, priority." Produce minimal output based on titles only (UNCLEAR_SCOPE check only).

---

Output Format

# Grooming Report — [date]

## Scorecard
- Total items: N
- Flagged items: N (X%)
- No owner: N
- No estimate: N
- Stale (>14 days): N
- Blocked without ETA: N
- Possible duplicates: N
- Priority inflation: YES / NO

## Top Issues for Grooming

| Item | Title | Flags | Priority |
|------|-------|-------|----------|
| [ID] | [Title] | NO_OWNER, NO_ESTIMATE | High |
| ... | ... | ... | ... |

## Grooming Agenda

### 1. [Topic]
**Items:** [ID1], [ID2]
**Problem:** [description]
**Recommended action:** [what to discuss / decide]

### 2. ...

## Skipped Checks
The following checks were not applied due to missing columns:
- [Flag name]: requires column "[column name]"

Field rules:

• Scorecard numbers must be accurate counts from the parsed data
• Top Issues table: max 10 rows, sorted by flag count descending
• Agenda topics: include only those with at least one affected item
• Skipped Checks: list only checks that were actually skipped; omit section if none skipped