Task-shaped repo context for coding agents

Map the repo before you burn context.

Cartograph is a CLI-first package that turns a repo into the smallest useful working artifact for the next agent.

Use it against local repos or GitHub repos. First map the repo, then build a typed task packet, then load only the context needed for the job. Claude Code, OpenClaw, MCP, and direct shell workflows all ride on the same core analysis engine.

CLI
first, MCP optional
2 skills
same output contract
1 package
Claude, OpenClaw, MCP
Zero to first result
npm install -g @anthony-maio/cartograph

cartograph analyze ./my-project --static
cartograph packet ./my-project --type bug-fix --task "fix auth"
cartograph context ./my-project --task "add auth" --json

/plugin marketplace add anthony-maio/cartograph
/plugin install cartograph@making-minds-tools

cartograph install claude
cartograph install openclaw
cartograph install mcp

Most repo handoffs waste tokens on the wrong files.

Dumping an entire codebase into context is slow, expensive, and noisy. Cartograph gives agents a smaller, better map: entry points, fan-in, API surfaces, dependency hubs, and task-scoped context chosen on purpose.

Without structure

Repo orientation becomes a context tax.

  • Agents read broad file trees just to get started.
  • Important wiring gets buried under leaf modules and tests.
  • Docs tasks pull too much parent-context prose.
  • Every new host integration rebuilds the same survey logic.
With Cartograph

The repo gets reduced to the useful shape.

  • CLI and MCP expose the same core analysis engine.
  • Artifacts land in user cache instead of dirtying the repo.
  • Claude Code and OpenClaw get packaged skills out of the box.
  • Downstream agents consume the same output contract either way.

Analyze, packet, context.

Cartograph works best when you treat it like a workflow, not a bag of commands.

01

Analyze

Map the repo, rank important files, and surface dependency hubs without dragging the whole tree into context.

02

Packet

Turn a concrete job like a bug fix or review into a reusable packet with likely edit points, risks, and validation targets.

03

Context

Load the smallest file set that supports the task you actually want to perform.

Install once, map the repo, then prepare the work.

This is the fastest useful path. You do not need MCP to start. Install the package, run the CLI, and only add host integrations where they help.

01

Install the package

Publish path is scoped. The executable remains cartograph.

npm
npm install -g @anthony-maio/cartograph
02

Analyze a repo

Start with static analysis to get ranked files and edges. Small repos stay compact by default; add --include-contents when you want embedded snippets.

local or GitHub
cartograph analyze ./my-project --static --json
cartograph analyze https://github.com/anthony-maio/cartograph --static --json
cartograph analyze ./my-project --static --json --include-contents
03

Build a task packet

Turn a concrete job into a reusable working artifact with key files, risks, and validation targets.

task packet
cartograph packet ./my-project --type bug-fix --task "fix auth refresh bug"
04

Pull task-scoped context

Ask for the smallest file set that supports the task you actually want to do.

task context
cartograph context ./my-project --task "add user authentication" --json
05

Install only the host you want

Use the Claude plugin marketplace or the hybrid CLI install model, depending on how you work.

Claude plugin or host installs
/plugin marketplace add anthony-maio/cartograph
/plugin install cartograph@making-minds-tools

cartograph install claude
cartograph install openclaw
cartograph install mcp

One package, three host paths.

Keep the core engine universal. Add host-specific behavior only where it helps.

Cartograph is also discoverable through the official MCP Registry as io.github.anthony-maio/cartograph.

Claude Code

Plugin, skills, and agent bundle

Install from this repo as a Claude plugin marketplace entry, or use the CLI host install if you prefer manual user-scope wiring.

  • /plugin marketplace add anthony-maio/cartograph
  • /plugin install cartograph@making-minds-tools
  • use-cartograph
  • repo-surveyor
  • repo-scout, dependency-tracer, context-picker, api-surface-writer, wiki-writer
OpenClaw

Skill pack plus OpenProse templates

Installs user-scope skills with bundled templates for repo docs and task-context flows.

  • use-cartograph
  • repo-surveyor
  • OpenProse templates for repo docs and task context
MCP

Optional server path

Use MCP only when the host wants tool wiring instead of shell calls.

  • analyze_repo
  • build_task_packet
  • get_file_contents
  • Reusable config snippet under the user cache

Tool-first when available. Manual fallback when not.

The point of shipping both skills is not redundancy. It is portability. Different acquisition path, same downstream contract.

use-cartograph

Use when the CLI or MCP server is available. It tells the host to run Cartograph directly for repo orientation, task context, documentation, and artifact export.

repo-surveyor

Use when Cartograph itself is unavailable or when you want a manual verification pass. It teaches the host how to replicate Cartograph behavior with targeted file discovery, ranking, dependency tracing, and synthesis.

Shared output contract

Downstream agents see the same shape either way.

key files
dependency hubs
minimal task context
doc-ready summary

Common first runs and public artifacts.

Cartograph now ships a public benchmark suite plus tracked example artifacts. Use the commands below to generate your own, then compare them to the linked `llama.cpp` packet and DeepWiki-style brief.

Local repo orientation
cartograph analyze ./my-project --static --json

Use this to get the first ranked map of a repo before touching code.

Task-scoped handoff
cartograph context ./my-project --task "trace the auth flow" --json

Use this when the next agent needs the smallest useful file set for a specific change.

Typed task packet
cartograph packet ./my-project --type bug-fix --task "fix auth refresh bug"

Use this when the next agent needs a reusable packet with likely edit points, exact validation targets, and shared dependency hubs instead of peripheral utility files.

Documentation output
cartograph wiki ./my-project --static

Use this to generate a structured docs pass without rereading the entire tree by hand.

Public benchmark suite
npm run benchmark:task-packets -- --list
npm run benchmark:task-packets -- --output benchmarks/task-packets/output

Use this to compare packet quality across `llama.cpp`, `vscode`, `next.js`, `fastapi`, `open-webui`, `kubernetes`, `home-assistant`, and `pytorch`.