Skill: investigation

What I do

I conduct systematic codebase investigations using parallel agent exploration, synthesise findings into a structured set of Obsidian documents, and create auto-generated DataviewJS indexes for discovery and navigation. The output is a reproducible, searchable investigation record stored in the user's Obsidian vault.

When to use me

• When conducting a discovery or audit of an application or codebase
• When asked to investigate, explore, or assess a project
• When producing structured findings for a codebase review
• When the user wants a documented record of a project investigation

Investigation Workflow

Phase 1: Plan

1. Identify project (name, language, entry point)
2. Identify vault (default: /home/baphled/vaults/baphled/)
3. Determine folder: 1. Projects/{Project}/Investigations/{YYYY-MM-DD}/
4. Create todo list to track progress

Phase 2: Explore (6 Parallel Agents)

Launch agents for: structure, architecture, debt, testing, CI/CD, documentation. Each returns metrics, file paths, and assessments.

Phase 3: Synthesise Documents

Create 6 numbered documents:

• 00-Executive-Summary.md — Good/Bad/Ugly, metrics, assessment
• 01-Architecture-Deep-Dive.md — Layers, patterns, violations
• 02-Technical-Debt-Analysis.md — Prioritised inventory
• 03-Testing-Strategy.md — Coverage, gaps, patterns
• 04-CI-CD-Assessment.md — Pipeline evaluation
• 05-Recommendations.md — Action plan

Phase 4: Create Auto-Generated Indexes

• Project-level: Investigations.md with DataviewJS auto-discovery
• Dated page: {YYYY-MM-DD}.md listing all documents

Phase 5: Store in Memory

Create memory entities for key findings.

Document Conventions

Frontmatter: Include title, date, type (discovery/investigation), project, status, created/modified timestamps.

Cross-linking: Use relative wikilinks (e.g., [[01-Architecture-Deep-Dive]]), not project-prefixed.

Tags: Add #investigation #project-slug #YYYY-MM-DD #discovery at bottom.

Numbering: 00-05 with kebab-case names (Executive-Summary, Architecture-Deep-Dive, etc.)

DataviewJS Rules

• ALWAYS use dv.table(headers, rows) for tables
• NEVER use dv.paragraph() with markdown table strings
• Project-level index: auto-discover dated folders, render status grid
• Dated page: list all documents with status

Folder Structure

1. Projects/{Project}/
  Investigations.md                 ← DataviewJS auto-index
  Investigations/{YYYY-MM-DD}.md    ← Dated page
  Investigations/{YYYY-MM-DD}/
    00-Executive-Summary.md
    01-Architecture-Deep-Dive.md
    02-Technical-Debt-Analysis.md
    03-Testing-Strategy.md
    04-CI-CD-Assessment.md
    05-Recommendations.md

Anti-patterns to avoid

• ❌ Hardcoding data in indexes — use DataviewJS auto-discovery
• ❌ Using dv.paragraph() for tables — use dv.table(headers, rows)
• ❌ Prefixing wikilinks with project name — keep relative
• ❌ Running agents sequentially — launch all 6 in parallel
• ❌ Skipping memory storage — store findings as entities
• ❌ Manual index files — must be auto-generated
• ❌ Forgetting frontmatter — required on all documents
• ❌ Mixing assessment with raw data — Executive Summary assesses only

KB Reference

~/vaults/baphled/3. Resources/Knowledge Base/AI Development System/Skills/Workflow-Orchestration/Investigation.md

Related skills

• research - General research methodology (investigation is a specialised form)
• obsidian-structure - PARA structure conventions for the vault
• obsidian-dataview-expert - DataviewJS queries and dashboards
• memory-keeper - Storing discoveries in the knowledge graph
• parallel-execution - Running exploration agents concurrently
• note-taking - General note creation conventions