codeintel-cli

Original：🇨🇳 Chinese

Translated

Use the codeintel command-line tool: intelligent indexing and field trace query for Go codebases. Features include building indexes (init), querying symbol/field read-write summaries (query fields), field tracing (trace-backward/forward), full-chain data value tracing (value-trace), call relationship analysis (callers/callees/impact), export (export), and Web service (serve). This skill is used when users request to analyze field data flows in Go codebases, trace field users/producers, check function field read-write operations, or query call relationships.

2installs

Sourceschaepher/codeintel

Added on2026-08-16

NPX Install

npx skill4agent add schaepher/codeintel codeintel-cli

SKILL.md Content (Chinese)

View Translation Comparison →

codeintel CLI User Guide

codeintel is an intelligent indexing tool for Go codebases (SSA field tracing), with output stored in SQLite (

.codeintel/codeintel.db

). Repository module:

github.com/schaepher/codeintel

.

Build and Installation

bash

# Build within the codeintel repository
go build -o codeintel ./cmd/codeintel
# Or install to GOBIN
make install

Command Quick Reference

bash

codeintel init --repo <path>               # Full index build (requires go.mod; will prompt to enter module directory at go.work root)
codeintel update --repo <path>             # Incremental update (git detects changed files, full analysis + incremental write)
codeintel serve --repo <path> --addr :8096 # Start graph exploration Web service (frontend AntV G6, default port :8090)
codeintel query <sub> ... --repo <path>    # Query (see below; all subcommands support --json / --compact)
codeintel export --repo <path> [--out x.json]  # Export field double-layer index JSON (field → producer/consumer)
codeintel export graph --type value-trace|callees|lifecycle --target <node> [--format mermaid|dot] [--out file]
                                           # Graph export: value-trace defaults to mermaid (function grouping), callees defaults to dot,
                                           # lifecycle diagram ([storage]/[observation]/[read]/[write] + condition annotations)
codeintel clean --repo <path> --force      # Delete index (must clean + init rebuild after schema changes)
codeintel version

query Subcommands

Subcommand	Purpose	Key Parameters
`symbol <sym>`	Symbol details (including callers/callees)
`fields <func>`	Function field read-write summary (direct_read/write + indirect_write)
`trace-backward <field> --func <func>`	Reverse trace of field production points	`--max-depth N` default 8
`trace-forward <field> --func <func>`	Forward trace of subsequent field usage	`--max-depth N`
`value-trace <nodeID>`	Full chain of data values (cross-function, grouped by function context)	`--max-depth N`
`callers/callees <sym>`	Callers/callees	`--depth N` default 1
`impact <sym>`	Impact analysis	`--depth N` default 3
`summary <node>`	Cross-layer summary: main chain of entry → compute → write → consume (each step with file:line)	`--format mermaid`

All query subcommands support
```
--json
```
(structured output) and
```
--compact
```
(no indentation)
Logs are written to
```
.codeintel/codeintel.log
```
(same directory as db), only query results are kept in stdout
```
<sym>
```
accepts canonical ID (
```
symbol:go:<pkg>:<name>
```
, method
```
(T).m
```
) or name (reports error and lists candidates if multiple matches)

<field>

is a type-qualified path (e.g.,

example.com/app/internal/agent.Config.APIKey

)

<nodeID>

is a complete node ID (e.g.,

symbol:go:...:(Manager).Run#m.cfg.APIKey.read@47

)

Real-world Examples

Take a Go repository (module

example.com/app

, with the

m.cfg.APIKey

field of an LLM agent) as an example:

bash

# 1. Build index (first time or after schema changes)
codeintel init --repo <target-repo>

# 2. Function field read-write summary (after verification, you can see [direct_read]/[direct_write]/[indirect_write] groups)
codeintel query fields "(Manager).Run" --repo <target-repo>

# 3. Forward trace of field users
codeintel query trace-forward example.com/app/internal/agent.Config.APIKey \
  --func "(Manager).Run" --repo <target-repo>

# 4. Full chain of data values (cross-function): first query node ID, then trace
sqlite3 <target-repo>/.codeintel/codeintel.db \
  "SELECT id FROM nodes WHERE kind='field_access' AND json_extract(properties,'\$.instance_path')='m.cfg.APIKey' LIMIT 1"
codeintel query value-trace "<ID obtained above>" --repo <target-repo>

Output Interpretation

fields summary:
```
[direct_read]
```
reads fields,
```
[direct_write]
```
writes fields,
```
[indirect_write]
```
writes indirectly via aliases/call closures (e.g.,
```
m.mu :55 m.mu.Lock()
```
). The summary table deduplicates by field UNIQUE (only the first line is listed for multiple accesses to the same field), details are in graph nodes
value-trace: grouped by
```
【function name】
```
+ indented tree,
```
←
```
production chain (reverse),
```
→
```
usage chain (forward), edge types (data_flows_to/argument/returns/phi_operand),
```
[read]/[write]
```
markers and line numbers
Intermediate layers of read chains (e.g., inner layer
```
m.cfg
```
of
```
m.cfg.APIKey
```
) are marked as read instead of write;
```
[]T{...}
```
literal initialization does not generate element nodes
Path conditions: trace lines may include
```
[condition: ...]
```
(if/type branch/env, calculated during query)
Dynamic dispatch: symbol interface types show candidate implementations (
```
[register 0.9]
```
/
```
[enum 0.7]
```
+ registration points)
Persistence: SQL writes are mapped to
```
users.name
```
virtual nodes (field → table.column, visible via value-trace)
Global traceability: global variables share nodes across functions (
```
var.<name>
```
), value-trace can reach initialization expressions
Cross-layer summary:
```
query summary <node>
```
outputs the main lifecycle chain (entry/compute/write/consume)

Verification and Notes

Change verification matrix:
```
make test
```
(unit),
```
make it
```
(integration, requires scip-go),
```
make e2e
```
(22 playwright cases, port 8096, specify verification repository with
```
E2E_REPO=<repo>
```
)
No automatic schema migration (user_version=2): after schema changes, verification repositories must run
```
clean --force
```
+
```
init
```
to rebuild, otherwise version mismatch error will occur
Push to git after each modification and verification (user agreement)
Logs: zap + OTel writes to
```
.codeintel/codeintel.log
```
(main rough parsing --repo passed to Setup); debug logs with --verbose are also in the file; stdout only contains query results
Pitfall:
```
pkill -f "codeintel-e2e serve"
```
will match its own command line and kill itself (use
```
pgrep -x codeintel-e2e
```
+ kill)
Index query has no network dependencies; build requires
```
go
```
and optional
```
scip-go
```
(scip adapter degrades and skips if missing)

codeintel-cli

NPX Install

Tags

SKILL.md Content (Chinese)

codeintel CLI User Guide

Build and Installation

Command Quick Reference

query Subcommands

Real-world Examples

Output Interpretation

Verification and Notes