Skip to content
Copilot Swarm

swarm analyze

Produces a detailed analysis of your repository covering tech stack, architecture, patterns, conventions, and a guide for implementing new features.

Usage

Section titled “Usage”
Terminal window
swarm analyze

Process

Section titled “Process”
  1. Architect exploration (primary model) — An architect agent explores the repository structure and produces a structured analysis.
  2. Senior engineer review (primary model) — Independently verifies accuracy and completeness (up to 3 iterations).
  3. Cross-model verification — The same loop with the review model, catching model-specific blind spots.

Large Repository Support

Section titled “Large Repository Support”

Repos with 500+ files automatically use chunked parallel analysis:

  1. Scout (fast model) — Scans structure and README for high-level overview.
  2. Partition — Groups directories into ~300-file chunks respecting natural boundaries.
  3. Parallel chunk analysis (primary model) — Multiple agents analyze chunks in parallel.
  4. Synthesis — Merges chunk analyses into a unified document.
  5. Review — Senior engineer reviews the synthesis.

Configure thresholds via environment variables:

  • ANALYZE_CHUNK_THRESHOLD — File count to trigger chunked mode (default: 500)
  • ANALYZE_CHUNK_MAX_FILES — Max files per chunk (default: 300)

Central Cache

Section titled “Central Cache”

Analysis results are cached centrally at ~/.config/copilot-swarm/analysis/<repo-key>/. Cache is keyed by git remote origin and validated via tree hash — same content = cache hit regardless of commit history.

Output

Section titled “Output”
.swarm/analysis/
  repo-analysis.md    # Final analysis document
  chunks/             # Individual chunk analyses (chunked mode only)