Retro Pattern Analyzer

Retro Pattern Analyzer

This skill analyzes retrospective files from multiple sprints and produces a structured report identifying recurring pain points, unresolved action items, and stable positive patterns. It reads .md and .txt files, normalizes different retro formats, and outputs retro-patterns-YYYY-MM-DD.md.

Input:

• 2 or more retrospective files in .md or .txt format
• Optional: folder path or list of file paths
• Optional: focus area (e.g., "technical issues only", "delivery problems")

Output:

• retro-patterns-YYYY-MM-DD.md — structured report with three blocks: recurring pains, unresolved action items, positive patterns

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

Instructions

Step 1: Validate Input

1. Check that at least 2 files are provided

   • If only 1 file provided: stop. Report: "Pattern analysis requires at least 2 retrospective files. Please provide a second file or folder path."
   • If folder path given: list all .md and .txt files in the folder; if fewer than 2 found, stop with same message

2. Check each file is readable

   • If any file is unreadable or path does not exist: skip that file, report: "File [name] not found or unreadable — excluded from analysis."
   • Continue with remaining files if at least 2 remain

3. If optional focus area provided: note it; use it to filter themes in Step 3

Step 2: Parse Retrospective Files

1. For each file, identify structural sections using keyword detection:

   • What went wrong / Problems / Improvements / Minuses / Keep Stop Start → went-wrong bucket
   • What went well / Strengths / Positives / Pluses → went-well bucket
   • Action items / Next steps / TODOs / Agreements → actions bucket
   • Russian equivalents: Что мешало / Проблемы / Минусы → went-wrong; Что помогло / Плюсы → went-well; Договорённости / Задачи → actions

2. Extract all bullet points or numbered items from each section

   • One item per line/bullet
   • Strip formatting markers (-, *, [ ], ✓)
   • Preserve the source file name as sprint identifier (use filename or date found in file header)

3. If file has no recognizable section headers:

   • Process entire file as free-text
   • Attempt keyword-based classification per line
   • Prepend note in report: "File [name] had no standard structure — classified by keywords"

4. Determine chronological order:

   • Use date in filename (e.g., retro-2026-03-15.md) if present
   • Else use date found in first line of file
   • Else use file modification date
   • Sort sprints chronologically from oldest to newest

Step 3: Identify Patterns

1. Normalize themes: Group items that describe the same issue using semantic similarity

   • Examples: "deployment takes too long" + "slow deploys" + "release process slow" → one theme: "slow release / deployment"
   • Aim for 5–12 distinct themes across all files; merge closely related items

2. Count theme frequency: For each theme, record which sprint files it appears in

   • Threshold for "recurring": appears in ≥2 sprint files

3. Calculate trend for each recurring theme:

   • ↑ growing — appears in 2+ consecutive recent sprints and not in early sprints
   • ↓ resolving — appeared in early sprints but not in the 2 most recent
   • → stable — appears consistently or non-consecutively without clear trend

4. Identify unresolved action items:

   • For each item in actions bucket of sprint N: check if the same or similar issue appears in went-wrong of sprint N+1 or later
   • If yes → mark as unresolved; record which sprint it was raised and which sprint it reappeared

5. Identify positive patterns:

   • Themes appearing in went-well across ≥2 sprints → stable positive pattern
   • Record frequency

6. Apply focus filter if provided: keep only themes matching the focus area keyword

Edge Cases:

• File with 50+ items: process all items, but group into max 10 themes plus an "Other" bucket for less frequent items
• Only 2 files (minimum set): process normally; add note in report: "Analysis based on 2 sprints — patterns are preliminary"
• Mixed languages (EN + RU files): detect per-file, normalize to report language (majority-language file wins)
• Multiple retro formats in one set: normalize to went-well/went-wrong/actions before pattern matching

Step 4: Generate Report

1. Use the output template structure (see Output Format below)
2. Fill three blocks:
   • Recurring Pains: themes from went-wrong with frequency ≥2, sorted by frequency descending
   • Unresolved Action Items: actions raised in one sprint that reappeared in a later sprint
   • Stable Positive Patterns: themes from went-well with frequency ≥2
3. If a block has no entries: write "None identified in this set of sprints"
4. Add metadata header: list sprint files analyzed, total item count processed, generation date

Step 5: Save Output

1. Write file as retro-patterns-YYYY-MM-DD.md (today's date)
2. Save to the folder where the retro files are located, or to the working directory if mixed paths were provided
3. Confirm: "Report saved as retro-patterns-[date].md — [N] themes identified across [N] sprints."

Output Format

# Retro Pattern Analysis
**Sprints analyzed:** [list of file names / date range]
**Total items processed:** [N] across [N] files
**Generated:** YYYY-MM-DD

---

## 🔴 Recurring Pains (went-wrong, ≥2 sprints)

| Theme | Sprints | Frequency | Trend |
|-------|---------|-----------|-------|
| [theme 1] | S1, S2, S4 | 3/4 | ↑ growing |
| [theme 2] | S2, S3 | 2/4 | → stable |

## 🔁 Unresolved Action Items

| Action Item | Raised in | Reappeared in | Status |
|-------------|-----------|---------------|--------|
| [action 1] | S2 | S3, S4 | unresolved |

## ✅ Stable Positive Patterns (went-well, ≥2 sprints)

| Theme | Sprints | Frequency |
|-------|---------|-----------|
| [theme] | S1, S3, S4 | 3/4 |

---
*Generated by retro-pattern-analyzer · [date]*

Field rules:

• Theme names: concise, 3–7 words, plain language
• Frequency format: N/total (e.g., 3/5 means appeared in 3 of 5 sprints)
• Trend: one of ↑ growing, ↓ resolving, → stable
• Sprint identifiers: use filename or inferred date label

Negative Cases

• Only 1 file provided → stop, ask for a second file. Do not produce partial report.
• File does not contain any recognizable retrospective content (no went-well/went-wrong/actions signals at all) → skip file, warn user, continue with remaining files.
• All provided files are unreadable or non-existent → stop. Report each missing path.
• Folder path does not exist → stop. Report the path and ask user to verify.