deep-investigation

name: deep-investigation description: Progressive iterative research with persistent knowledge vault, entity tracking, and coverage gates. USE WHEN deep investigation, investigate, deep research, map the landscape, competitive analysis, market mapping, technology survey, thorough research.

Deep Investigation

Mode: Progressive iterative research with persistent knowledge vault Time: 10-60 minutes | Can span multiple sessions via persistent vault files

When to Use

• User says "deep investigation", "investigate [topic]", "deep research", "map the [X] landscape"
• Competitive analysis, market mapping, technology surveys, threat landscape analysis
• Any research needing iterative deepening -- broad discovery first, then progressive deep dives
• When surface-level research is insufficient and systematic coverage matters

Concept

Progressive narrowing funnel:

Iteration 1: Broad landscape --> discover entities --> score importance --> deep-dive top entity
Iteration 2: Read vault artifacts --> pick next highest-priority entity --> deep-dive
Iteration 3+: Continue until all critical entities researched and all categories covered

Single-run: One full cycle (landscape through first deep dive). Multi-session: User returns later; workflow detects existing vault state and continues.

Available Tools

Vault Location

~/.augment/MEMORY/RESEARCH/{YYYY-MM-DD}_{topic-slug}/

Vault survives across sessions. Workflow is stateless -- all state lives in files on disk.

Workflow

Step 0: Detect Iteration State

Check vault directory for existing artifacts:

• LANDSCAPE.md + ENTITIES.md exist --> CONTINUATION: read ENTITIES.md
  • PENDING CRITICAL/HIGH entities? --> Step 4 (Investigate)
  • All CRITICAL/HIGH done but thin categories? --> Step 3 (Discover)
  • All gates pass? --> Step 5 (Summarize)
• Neither exists --> FIRST ITERATION: start at Step 1

Step 1: Landscape Scan (First Iteration Only)

Goal: Map the full landscape. Do this once, reference cheaply in all later iterations.

Run 4-6 WebSearch queries covering: domain overview, key players, recent developments, historical context, adjacent domains, contrarian views. Deep-read top 3-4 results with WebFetch.

Produce three vault files:

LANDSCAPE.md -- Domain synthesis with sections: Overview (2-3 paragraphs), Domain Structure, Key Dynamics, Entity Categories (discovered dynamically, with estimated counts), Initial Entity Discoveries, Sources (verified URLs only).

ENTITIES.md -- Master catalog table with columns: Entity, Category, Status (PENDING/RESEARCHED/SKIP), Value (CRITICAL/HIGH/MEDIUM/LOW), Effort (EASY/MODERATE/HARD), Profile link. Include legend definitions at the top.

INDEX.md -- Navigation hub linking to LANDSCAPE.md, ENTITIES.md, and profiles as they are created. Track coverage stats: categories complete, entities researched, CRITICAL/HIGH pending.

Step 2: Evaluate (Score Entities)

For each PENDING entity, assess two dimensions:

VALUE: CRITICAL (category definers, must research) > HIGH (major players) > MEDIUM (notable) > LOW (minor) EFFORT: EASY (abundant public info) > MODERATE (some digging) > HARD (limited info)

Investigation priority: CRITICAL+EASY first, then CRITICAL+HARD, HIGH+EASY, HIGH+HARD. MEDIUM+ only after all CRITICAL/HIGH done. Update ENTITIES.md with scores.

Step 3: Discover (Expand Coverage)

Find entities in categories with fewer than 3 entries. For each thin category:

WebSearch(query: "notable {category} in {domain} space")
WebSearch(query: "{domain} {category} key players landscape")

WebFetch the best results. Add discoveries to ENTITIES.md as PENDING, then score them (Step 2).

Step 4: Investigate (Deep Dive -- One Entity)

Quality over quantity. Research ONE entity per iteration.

Select the highest-priority PENDING entity (VALUE desc, EFFORT asc). Run 2-3 focused WebSearch queries:

• "[entity] overview analysis {domain}"
• "[entity] recent news funding developments"
• "[entity] vs competitors strengths weaknesses"

Deep-read top 2-3 results with WebFetch.

Save profile to vault: {Category}/{entity-slug}.md containing: overview, key facts (founding, size, funding, leadership), products, competitive advantages/weaknesses, recent developments, cross-links to related vault entities, verified source URLs.

Update ENTITIES.md (mark RESEARCHED, add link) and INDEX.md (add to navigation).

Step 5: Progress Check (Coverage Gates)

Two gates must BOTH pass for completion:

Breadth Gate: Every entity category has >= 3 entities (excluding SKIP). Depth Gate: Every CRITICAL or HIGH entity is RESEARCHED or SKIP.

Both pass: Produce SUMMARY.md (executive synthesis), update INDEX.md with final stats, report completion. Either fails: Report current coverage, identify gaps (thin categories, pending CRITICAL/HIGH entities), ask user whether to continue or stop.

Output Artifacts

{vault}/
  INDEX.md          -- Navigation hub with coverage stats
  LANDSCAPE.md      -- Broad domain analysis
  ENTITIES.md       -- Master catalog with status tracking
  SUMMARY.md        -- Executive synthesis (on completion)
  {Category}/       -- Entity profiles by category
    entity-a.md
    entity-b.md

All profiles cross-linked. Vault is self-contained and readable standalone.

Single-Run vs Multi-Session

Single-run: complete Steps 1-4, report what remains PENDING.

URL Verification

All vault URLs MUST be verified via WebFetch before inclusion. A broken link in a persistent vault is a compounding error -- it misleads every future session that reads it.