stacklit
llms.txt131 starsChanged 7 days agoMITView on GitHub
Stacklit writes a small, committable JSON index of a codebase (modules, exports, dependency edges, hints) and a symbol graph, so AI coding agents can orient themselves without re-exploring the repo each session. Free, MIT licensed, one Go binary, deterministic, runs locally with no LLM, and the CLI makes no network calls. Key facts: - Install: npx stacklit init, brew install glincker/tap/stacklit, npm install -g stacklit, or go install github.com/glincker/stacklit/cmd/stacklit@latest. - Output: stacklit.json (index), a Mermaid dependency file (DEPENDENCIES.md by default,…
- Installs packages
The file
We split this llms.txt into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 14
Not a search engine. When the exact identifier is known, grep is cheaper.
≈ “Common” is an estimate: at least 1 in 5 of the 3,592 llms.txt files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
# Stacklit > Stacklit writes a small, committable JSON index of a codebase (modules, exports, dependency edges, hints) and a symbol graph, so AI coding agents can orient themselves without re-exploring the repo each session. Free, MIT licensed, one Go binary, deterministic, runs locally with no LLM, and the CLI makes no network calls. Key facts: - Install: `npx stacklit init`, `brew install glincker/tap/stacklit`, `npm install -g stacklit`, or `go install github.com/glincker/stacklit/cmd/stacklit@latest`. - Output: `stacklit.json` (index), a Mermaid dependency file (`DEPENDENCIES.md` by default, `output.mermaid` in `.stacklitrc.json`), `stacklit.html` (interactive map), `stacklit derive` (compact text map for prompts). - Deterministic: same repo gives the same bytes. Hard token budgets on queries (`--budget`). - Languages: 48 languages and formats in the index; the symbol graph (query, explain, impact, path, hubs, slice) covers 20 languages. - Assistants: `stacklit setup` configures 26 targets, including Claude Code, Cursor, Aider, Codex, Gemini CLI, GitHub Copilot, Windsurf, Cline, Zed and a generic AGENTS.md. MCP server via `stacklit serve` (stdio, 15 tools, or `--profile minimal` for 4). - Privacy: no telemetry, no account, no upload. The generated HTML map requests small language logos from thesvg.org unless `STACKLIT_MAP_ICONS=off`. The only opt-in external call is `stacklit init --summary`. - Benchmarks: reproducible, no LLM calls, six pinned repos. The derived map is 137 to 1,351 tokens. The benchmark does not show a token saving over grep; it shows cheap orientation. Limitations are documented. - Agent test: two rounds of real Claude Code sessions (450, then 650) on public repos. Stacklit cut tool calls by a quarter to a third but did not lower cost overall. Round 1: pooled cost ratio 1.00 (0.93 to 1.07), cheaper on the largest repo (0.78), more expensive on the two small ones (1.10, 1.23), answers right slightly less often. Round 2, four repos: largest repo 0.89 (0.84 to 0.96), smaller repos 1.05 to 1.61, pooled 1.09 to 1.17, so it cost more on average; answers equally right. See bench/agent-ab/RESULTS.md. - Not a search engine. When the exact identifier is known, grep is cheaper. ## Docs - [README](https://github.com/glincker/stacklit/blob/master/README.md): overview and install - [Usage](https://github.com/glincker/stacklit/blob/master/docs/USAGE.md): commands, MCP tools, config, index schema - [Install](https://github.com/glincker/stacklit/blob/master/docs/INSTALL.md): every install path, Claude Code plugin, Docker, MCP registry - [Benchmarks](https://github.com/glincker/stacklit/blob/master/bench/RESULTS.md): methodology, results, limitations - [Languages](https://github.com/glincker/stacklit/blob/master/docs/LANGUAGES.md): what is extracted per language - [FAQ](https://github.com/glincker/stacklit/blob/master/docs/FAQ.md) - [Privacy](https://github.com/glincker/stacklit/blob/master/docs/PRIVACY.md) - [Architecture](https://github.com/glincker/stacklit/blob/master/docs/ARCHITECTURE.md) - [Skill for agents](https://github.com/glincker/stacklit/blob/master/skills/stacklit/SKILL.md) ## Optional - [Full docs in one file](https://stacklit.glinr.com/llms-full.txt) - [Comparison](https://github.com/glincker/stacklit/blob/master/docs/COMPARISON.md) - [Roadmap](https://github.com/glincker/stacklit/blob/master/docs/ROADMAP.md) - [Changelog](https://github.com/glincker/stacklit/blob/master/CHANGELOG.md)
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.