Skip to content
repo

code-intelligence

Find out what is actually true about a codebase before you change it

1 skills 2 human gates 5–20 min/session
Install this pack Read the reference

The contract

Use it when
You need to know what calls something, what a change would break, how an unfamiliar system is organised, or where a behaviour is implemented — and you want the answer from a resolved call graph rather than a text search.
You type
What breaks if I change the request handler?
You provide
A repository indexed with `wicked-estate index .`, and the symbol, module, or behaviour you are asking about. Nothing else — the pack never writes to the graph without asking.
You receive
An answer grounded in the index, carrying the index's own limits: how many references it could not resolve, whether output continues on later pages, and which revision it describes. Where the graph cannot answer, you are told that rather than given a guess.

Where you decide

The agent pauses at these points. You choose whether to continue, redirect, or stop.

Confirm which symbol you meantDecision 1 of 2Approve building the indexDecision 2 of 2

What changes when you install this

Before, an agent answering "what depends on this?" grepped for a name and produced a list it could not vouch for — no way to tell a real call from a coincidence, and no way to say what it missed. After installing this pack, the same question is answered from resolved call and import edges, and the answer states what the indexer could not bind. The habit matters more than the tool: resolve the symbol before reading it, read the source before concluding, stop when the question is answered, and never present a text search as a blast radius.

The journey

Code intelligence

Point an agent at a question about your codebase and get an answer it can show its working for.

The pack drives the Wicked Estate command-line tool against an index you build. It is read-only by default: the commands that write to the graph are named, and each one asks first.

What it will not do is guess. If the index cannot answer — and the pack ships a fourteen-point assessment of exactly what it cannot — you are told so, with what it can tell you instead.

The journey

What breaks if I change the request handler?

1. Get an index

  • You provide: a repository, and cargo on your PATH.
  • Agent does: checks for the binary and an existing index, and reports which is missing rather than guessing.
  • You decide: approve building the index, which writes a few hundred megabytes into the working tree, and confirm .wicked-estate/ is ignored.
  • Output: a graph of the repository, with its node, edge and file counts.
  • State: confirmed-write

2. Resolve what you asked about

  • You provide: a symbol, module, or behaviour, in ordinary language.
  • Agent does: resolves the name to stable symbol IDs and reports every match with its file and line.
  • You decide: which symbol you meant, when the name is ambiguous.
  • Output: one chosen symbol, with the choice’s reason stated.
  • State: read-only

3. Gather the evidence

  • Agent does: selects the investigation pattern the question calls for, reads source before concluding, expands one hop at a time, and stops when the question is answered rather than when the graph is exhausted.
  • Output: the findings, separating what the graph showed from what was verified against source.
  • State: read-only

4. Read the limits with the answer

  • Agent does: reports the index’s own completeness — references it could not bind, output still on later pages, the revision it describes — and names what it could not establish.
  • You decide: whether the evidence is enough for your purpose, or a narrower question is needed.
  • Output: an answer you can act on, bounded by what the index can support.
  • State: read-only

Human gates

For each gate, everything you need to make a confident decision.

  • Confirm which symbol you meant

    Trigger
    When a name resolves to more than one symbol, before any analysis runs on it
    Time
    under a minute
    What to check, good, bad, consequence

    What to check

    • Does the file path of the chosen match point at the code you had in mind?
    • If two matches look plausible, is one of them a vendored or generated copy?
    • Did the agent pick for a stated reason, or silently take the first hit?

    What good looks like

    Every match listed with its file and line, and either a question to you or a named reason for the choice.

    What bad looks like

    One symbol analysed with no mention that the name was ambiguous.

    Consequence of skipping

    Impact analysis on the wrong overload is worse than none — it reads as evidence while describing different code.

  • Approve building the index

    Trigger
    Before the agent runs `wicked-estate index`, which writes into your working tree
    Time
    under a minute, then several minutes unattended
    What to check, good, bad, consequence

    What to check

    • Is `.wicked-estate/` ignored by version control? The database is a few hundred megabytes.
    • Are you indexing the repository you meant, and not a parent directory?
    • If an index already exists, is rebuilding it what you want, or is the existing one fine?

    What good looks like

    The target path stated, the write disclosed, and the gitignore checked before the command runs.

    What bad looks like

    A large database appearing in your working tree, or committed, because nobody mentioned it.

    Consequence of skipping

    An un-ignored index dirties every status check and can be committed by accident; a stale one silently answers questions about an older revision.

Typical session

Agent turns
3–8
Human gates
2
Wall-clock time
5–20 min

Install

agentbundle install --pack code-intelligence

Skills in this pack

  • code-intelligence1 gate

    Query an indexed code graph through five reusable investigation patterns — understand an entity, analyze change impact, investigate behavior, analyze architecture, assemble task context — preserving the provenance, confidence and completeness the index reports. Read-only; degrades to labelled repository search when no index is present.

Technical reference & next steps