Compatibility: The context: fork frontmatter field is Claude Code-specific — on OpenCode and VS Code Copilot it is ignored and the skill runs inline using the current model. Core functionality is preserved on all platforms.Generate architecture documentation from a codebase scan. Writes to docs/architecture/.
Step 0: Parse Flags
Extract flags from the argument string:
BG = true if --bg present
SILENT = true if --silent present
OUT_DIR = docs/architecture/Create the output directory if it does not exist:
mkdir -p docs/architecture/Step 1: Parallel Codebase Scan
Launch four background agents to scan the codebase in parallel.
Agent A -- Component Map
model: claude-haiku-4-5-20251001
run_in_background: trueInstructions for Agent A:
Glob the repo for top-level directories, key entry points, plugin/package boundaries. Identify major subsystems and what each owns. For each component, note:
- Directory path
- Purpose (1 sentence)
- Key files / entry points
- What it exports or exposes to other components
Use Glob and Read to inspect the codebase structure. Focus on directories like plugins/, docs/, scripts/, src/, lib/, and any package/module boundaries. Return a structured list of components.
Agent B -- Data Flows
model: claude-haiku-4-5-20251001
run_in_background: trueInstructions for Agent B:
Trace how data moves between components -- API calls, file writes, event emissions. Focus on boundaries (what crosses from one component to another). Look for:
- Inter-plugin invocations (one skill calling another)
- File I/O patterns (which components read/write which directories)
- Script invocations across plugin boundaries
- Any HTTP/API calls between components
Use Grep and Read to find cross-boundary references. Return a list of data flows with source, destination, and mechanism.
Agent C -- API Inventory
model: claude-haiku-4-5-20251001
run_in_background: trueInstructions for Agent C:
Find all public interfaces -- HTTP routes, exported functions, CLI commands, skill invocation points. List with signatures. Specifically:
- All skill SKILL.md files (these are the primary public interface)
- Any exported scripts in
scripts/directories - CLI entry points
- Configuration schemas (plugin.json files)
Use Glob to find all SKILL.md and plugin.json files, then Read key ones. Return a table of interfaces with name, type, location, and brief description.
Agent D -- Dependency Graph
model: claude-haiku-4-5-20251001
run_in_background: trueInstructions for Agent D:
Parse package.json, pyproject.toml, go.mod, or similar files for external dependencies. Also map inter-plugin dependencies in this repo:
- Which plugins reference scripts from other plugins
- Which skills invoke other skills
- Shared utilities and who uses them
Use Glob to find dependency manifests and Grep to find cross-references between plugins. Return external dependencies and an internal dependency map.
Wait for all four agents to complete. Collect their results.
Step 2: Generate Documentation Files
Using the combined results from all four agents, generate the following files:
docs/architecture/README.md
Overview document:
- What the system is (1-2 paragraphs)
- Major components at a glance (bulleted list)
- How to navigate these docs (links to other files)
docs/architecture/components.md
Component map:
- ASCII diagram showing subsystem boundaries and ownership
- For each component: path, purpose, key files, what it owns
- Group by layer (plugins, docs, config, scripts)
docs/architecture/data-flows.md
Data flow documentation:
- Key data flows as ASCII sequence or flow diagrams
- Cross-boundary interactions
- File I/O patterns
- Focus on the most important 5-10 flows
docs/architecture/api-inventory.md
API inventory table:
- All public interfaces (skills, scripts, CLI commands)
- Columns: Name, Type, Location, Description
- Grouped by plugin
docs/architecture/dependencies.md
Dependency documentation:
- External dependencies (from package manifests)
- Internal dependency graph (inter-plugin references)
- ASCII diagram of internal dependencies
Step 3: Write Files
Write all five documentation files using the Write tool.
If BG is true and SILENT is not true, send a completion notification:
npx @codevoyant/agent-kit notify --title "dev:docs complete" --message "Architecture docs written to docs/architecture/"Report the files written and their sizes.