Sprang analyze

Analyze the codebase and produce .sprang/knowledge-graph.json with full semantic enrichment. You are the analysis engine — read every file, write rich summaries, detect architecture, score risk.

CRITICAL: Complete ALL 8 phases (Phase 0 through Phase 7) in one run. Stopping early leaves the Architecture, Domains, and Learn tabs empty. NEVER write .sprang/knowledge-graph.json yourself — Phase 7 assembles it. RESUME: If a graph already exists at phase: complete, jump to Phase 4 to re-run enrichment only.

For the full phase-by-phase procedure, read REFERENCE.md in this skill's directory. Read it before starting — the summary below is not sufficient to execute the skill correctly.

Options

• --full — force complete rebuild even if a graph exists
• --language <lang> — output summaries in a specific language (ISO code: zh, ja, ko, es, fr, de, pt, ru)
• --chunk <N> — split node output into chunks of N nodes
• A directory path — analyze that directory instead of cwd

Phases

1. Pre-flight — resolve PROJECT_ROOT and SPRANG_ROOT="$PROJECT_ROOT/.sprang", read .sprangignore, decide incremental vs full, collect README/manifest context and the entry point.
2. Scan — enumerate files, detect languages/frameworks, build the project-internal import map → write scan-result.json.
3. Analyze files — semantic batching (related files together), max 10 files and ~800 lines per batch → write final-nodes-chunk-*.json (max 50 nodes each), final-edges.json, and assembled-graph.json (project metadata only).
4. Architecture layers — assign every file node to exactly one of 3-10 layers → write final-layers.json directly (not layers.json).
5. Guided tour — 5-8 BFS-ordered steps → write final-tours.json as an array of Tour objects (not a flat step array — each needs id, title, description, steps).
6. Domain mapping — cluster into 2-6 business domains → write final-domains.json with domain/flow/step structure.
7. Risk + smells + git layer → write risk-scores.json as {"<node-id>": {"risk_score": 0.0, "risk_factors": [], "structural_warnings": [], "decision_context": {...}}}. The assemble step applies all of these fields to the nodes.
8. Assemble — prefer the CLI sprang merge "$PROJECT_ROOT" (normalises + validates, no Python needed). If sprang is not on PATH, fall back to PROJECT_ROOT="$PROJECT_ROOT" python3 skills/sprang-analyze/scripts/merge.py. Then write .sprang/SPRANG_REPORT.md.

All intermediate files live in .sprang/intermediate/ and both readers consume these exact names: final-nodes-chunk-*.json, final-edges.json, final-layers.json, final-tours.json, final-domains.json, risk-scores.json, assembled-graph.json. All must exist before assembling.

Schema reminders

• Do NOT emit a layer field on nodes — layer membership comes from final-layers.json. A layer of null fails validation.
• Use only canonical edge types: imports, exports, contains, inherits, implements, calls, subscribes, publishes, middleware, reads_from, writes_to, transforms, validates, depends_on, tested_by, configures, related, similar_to, deploys, serves, provisions, triggers, migrates, documents, routes, defines_schema, contains_flow, flow_step, cross_domain, cites, contradicts, builds_on, exemplifies, categorized_under, authored_by.
• Test coverage is tested_by: source file → the test file that covers it (note the direction: source → test).

After completion

Report files analyzed, nodes/edges, top risks, layers and domains found. Suggest the sprang-chat skill to ask questions, sprang-onboard for the guided tour, and pnpm --filter @sprang/dashboard dev to open the dashboard.