index-knowledge_skill

This skill generates a hierarchical AGENTS.md knowledge base for a codebase, creating root and subdirectory documentation with complexity scoring.

Rust

17.4k

GitHub Stars

1

Bundled Files

2 months ago

Catalog Refreshed

4 months ago

First Indexed

Readme & install

Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.

Installation

Preview and clipboard use veilstrat where the catalogue uses aiagentskills.

npx veilstrat add skill tursodatabase/turso --skill index-knowledge

SKILL.md9.8 KB

Overview

This skill generates a hierarchical AGENTS.md knowledge base for a codebase, producing a root file plus complexity-scored subdirectory documents. It detects directories that warrant their own AGENTS.md, scores them by code and structural signals, and writes concise, non-redundant docs in parallel. Use update or create-new modes to preserve or regenerate existing content.

How this skill works

The tool discovers project structure using parallel explore agents, bash probes, and LSP codemaps where available, then merges findings with any existing AGENTS.md content. It scores directories using a weighted matrix (file counts, symbol density, exports, references, etc.) and selects locations to document. Finally it generates a root AGENTS.md and per-directory AGENTS.md files in parallel, then deduplicates and trims to meet quality gates.

When to use it

Onboarding new contributors to quickly surface project structure and conventions
Before major refactors to map module boundaries and hotspots
When a repo lacks centralized, context-aware documentation
To maintain up-to-date docs after large merges or architecture changes
For multi-package repos where some folders deserve focused knowledge files

Best practices

Always read existing AGENTS.md files first; preserve useful context before regenerating
Prefer concise, project-specific notes — avoid generic advice and duplicates from parent files
Run discovery with LSP when possible; fall back to AST-grep and bash probes if unavailable
Tune max-depth and spawn thresholds for large repos to control doc granularity
Generate root first, then subdir files in parallel and run a final dedupe/trim pass

Example use cases

Generate a root AGENTS.md for a Rust-based embedded SQL engine and add AGENTS.md for high-complexity modules (storage, sql parser)
Run in update mode to refresh docs after introducing WebAssembly bindings or new API layers
Create-new mode to rebuild the entire knowledge base when structure was heavily reorganized
Identify deep modules with many symbols and produce focused AGENTS.md for them without repeating root content
Spawn additional explore agents automatically for big repos (many files/lines/depth) to uncover hidden hotspots

FAQ

By default it updates existing files and creates new ones where warranted. Use --create-new to read, then remove and regenerate all files.

How does the tool decide which directories get AGENTS.md?

It uses a weighted scoring matrix (file count, subdirs, symbol density, exports, references, unique configs) and thresholds to classify directories as root, high complexity, distinct domain, or skip.

Can it run without LSP?

Yes. LSP improves symbol and reference accuracy, but the discovery phase falls back to bash analysis and AST-grep when LSP is unavailable.