Skip to content
agent-ready-repo
GitHub

Code Intelligence

Your agent can answer what is true about your code, such as what calls this and what breaks if you change it, by querying an index instead of guessing.

Install

Section titled “Install”

You need agentbundle first. Install it from inside the repository you want to query:

cd path/to/your-repo
agentbundle install --pack code-intelligence

The pack drives Wicked Estate, which you install and run once to build the index:

cargo install wicked-estate --version 0.30.0 --locked
echo '.wicked-estate/' >> .gitignore
wicked-estate index .

It also installs at user scope:

claude plugin install code-intelligence@agent-ready-repo

What your agent can do

Section titled “What your agent can do”
  • Find a symbol and explain a class or service from the code graph.
  • Work out the impact of a change: who calls it, what depends on it, and which dependents are direct or indirect.
  • Trace behavior and value lineage through the code.
  • Spot hotspots and clusters in the architecture.
  • Gather the context a task needs, and say how many call sites the index could not resolve.

Skills

Section titled “Skills”
  • code-intelligence — Maps each question to the right Wicked Estate command, tells you when to stop, and keeps a guessed link from being reported as a fact.
  • code-investigator — A subagent that investigates how or why something happens in a fresh context. It labels each finding observed, verified, or unestablished.
  • impact-analyst — A subagent that analyses the impact of changing one symbol, module, or file, and states its own limits first.

Works with

Section titled “Works with”

Works with Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Kiro IDE, and Kiro CLI. Install Core too, which is optional: when both are installed, Core’s exploration skill can use this pack for code-graph evidence. See Understand the code.

Questions

Section titled “Questions”

How do I check my setup? Ask your agent to check the code-intelligence setup. It reports whether the binary, the index, and the version are ready; see Setup check.

What if there is no index? The agent falls back to ordinary repository search and labels it as search. A list of callers from grep is not a blast radius.

Does it change my code? No. It reads. Commands that change the graph, such as index and annotate, always ask first.

Details

Section titled “Details”

Scope: repo (default), user · Version: 0.1.10 · Source: packs/code-intelligence

Setup check

Section titled “Setup check”

Run python '<skill-dir>/scripts/estate_preflight.py' --check. <skill-dir> is the installed code-intelligence skill folder, such as .claude/skills/code-intelligence at repo scope. Exit 0 is ready, 2 means the binary is missing, 3 means no index, and 4 means the binary is older than the 0.30 floor.

More detail

Section titled “More detail”

Why does an empty answer from the MCP server not mean my code is empty? With --readonly, a store path that does not exist is served as an empty in-memory store. Check --db, WICKED_ESTATE_DB, or WICKED_MEMORY_DB first. By default the pack uses the CLI, not MCP.

Must another provider match Wicked Estate? No. The composition example is illustrative, not a contract. Other providers may expose fewer, different, or new capabilities and need not emulate Wicked Estate. The current investigation patterns may change.