dd-debugger

★ 940

by datadog-labs · part of datadog-labs/pup

Live Debugger - inspect runtime argument/variable values in production by placing log probes on methods. Use when asked what values a function receives, what parameters look like at runtime, or to capture live data from running services without redeploying.

🔌 This skill ships inside the pup plugin — install the plugin and you also get 40 sub-agents.

This is the playbook your agent receives when the skill activates — you don't need to read it to use the skill, but it's here to audit before installing.

Datadog Live Debugger

Place log probes on running services without redeploying. Create probes with custom templates and conditions, and stream captured events in real time.

Authentication

Authenticate via OAuth2 (recommended) or API keys:

# OAuth2 (recommended)
pup auth login

# Or use API keys
export DD_API_KEY="key" DD_APP_KEY="key" DD_SITE="datadoghq.com"

Typical Workflow

Prefer --capture expressions over full snapshots. Capture expressions are lighter-weight, faster, and return exactly the data you need.

If you don't know the service, ask the user before proceeding.
Verify the service using pup debugger context <service> to list environments with active instances. If multiple environments exist, ask the user which one to target before proceeding.
Find a method using the dd-symdb skill (pup symdb search --view probe-locations)
Place a probe with capture expressions for the values you need
Watch events with --fields for compact output
Delete the probe when done

# 0. List environments (if multiple, ask user which to use)
pup debugger context my-service --fields service,language,envs

# 1. Create a probe with capture expressions (recommended)
#    Use --fields id to get just the probe ID back
pup debugger probes create \
  --service my-service \
  --env production \
  --probe-location "com.example.MyController:handleRequest" \
  --capture "request.id" \
  --capture "request.headers" \
  --ttl 1h \
  --fields id

# 2. Stream events with compact output
pup debugger probes watch <PROBE_ID> --timeout 60 --limit 10 \
  --fields "message,captures,timestamp"

# 3. Clean up
pup debugger probes delete <PROBE_ID>

Service Context

Always run this before creating probes. It returns JSON by default (like all other commands) showing environments with active instances, tracer versions, and supported probe features. If the service runs in multiple environments, ask the user which one to target — don't guess.

# Full JSON output (default)
pup debugger context my-service

# Compact: just the fields you need
pup debugger context my-service --fields service,language,envs

# Filter to a specific environment
pup debugger context my-service --env production --fields service,envs,repo

Flag	Description	Default
`--env`	Filter to a specific environment	All environments
`--fields`	Comma-separated fields: `service`, `language`, `envs`, `repo`	Full JSON response

Probe Management

List Probes

# All probes
pup debugger probes list

# Filter by service
pup debugger probes list --service my-service

Get Probe Details

pup debugger probes get <PROBE_ID>

Create a Log Probe

pup debugger probes create \
  --service my-service \
  --env staging \
  --probe-location "com.example.MyClass:myMethod" \
  --capture "user.name" \
  --capture "order.items[0].price"

To disambiguate overloaded methods, pass a signature with argument types:

pup debugger probes create \
  --service my-service \
  --env staging \
  --probe-location "com.example.MyClass:myMethod(int, java.lang.String)" \
  --capture "user.name"

Options:

Flag	Description	Default
`--service`	Service name (required)	—
`--env`	Environment (required)	—
`--probe-location`	`TYPE:METHOD` or `TYPE:METHOD(args)` (required). The signature form disambiguates overloaded methods.	—
`--language`	`java`, `python`, `dotnet`, `go`	Auto-detected from symdb
`--capture EXPR`	Capture expression (repeatable). Use dot notation for fields, brackets for indexing.	None
`--capture`	Without value: enable full snapshot (capture everything).	No snapshot
`--template`	Log message template with `{variable}` placeholders. Can combine with `--capture`.	Auto-generated
`--condition`	DSL condition to filter captures	None
`--depth`	How deep the tracer traverses the object graph when capturing (1–5). Start at 1 to see field names/types, increase to drill in.	`1`
`--rate`	Snapshots per second	`1`
`--budget`	Max probe hits. Only "total" window supported (hourly/daily not yet available).	`1000`
`--ttl`	Probe time-to-live (e.g., `10m`, `1h`, `24h`). Probe auto-expires.	`1h`

Capture Expressions (Recommended)

Always prefer capture expressions over full snapshots. They are lighter-weight and return exactly the data you need.

# Capture specific fields using dot notation
--capture "user.name"
--capture "request.headers"

# Access array elements
--capture "orders[0].total"
--capture "items[len(items)].name"

# Chain member access
--capture "response.body.data.id"

Multiple --capture flags can be combined. Each expression is independently evaluated.

# Capture multiple specific values
pup debugger probes create \
  --service my-service --env prod \
  --probe-location "OrderService:processOrder" \
  --capture "order.id" \
  --capture "order.items[0].price" \
  --capture "customer.email"

Tip: Start with --depth 1 (the default) to see field names and types, then increase depth to drill into interesting subtrees. Use --depth 5 for deeply nested objects.

Full Snapshot (use sparingly)

Use bare --capture (no value) only when you don't know which fields to inspect:

# Full snapshot — captures all arguments, locals, return value
pup debugger probes create \
  --service my-service --env staging \
  --probe-location "com.example.MyClass:myMethod" \
  --capture

You can combine snapshot with capture expressions:

# Full snapshot + specific expressions
--capture --capture "user.name"

Templates

Templates can be used alongside capture expressions for custom log messages:

# Template with capture expressions
pup debugger probes create \
  --service my-service --env prod \
  --probe-location "MyClass:myMethod" \
  --capture "user.id" \
  --template "Processing request for user={user.id}, took {@duration}ms"

Template syntax uses {variable} placeholders:

--template "Processing order={orderId} for user={userId}"
--template "handleRequest took {@duration}ms"

Conditions filter when the probe fires:

--condition "status == 'error'"
--condition "@duration > 100"

Delete a Probe

pup debugger probes delete <PROBE_ID>

Watch Probe Events

Stream log events and status errors from a probe in real time.

pup debugger probes watch <PROBE_ID>

Options:

Flag	Description	Default
`--timeout`	Exit after N seconds	`120`
`--limit`	Exit after N log events	unlimited
`--from`	Start time for log query	`now`
`--wait`	Wait up to N seconds for the probe to become available	`0`
`--fields`	Comma-separated fields to include: `message`, `captures`, `timestamp`. Compact JSON output.	Full debugger payload

Behavior:

Without --fields: outputs the trimmed debugger payload (snapshot, probe info, stack) — not the full log envelope
With --fields: outputs only the requested fields as compact JSON per event
Polls for new log events and probe status errors every 1s
Default --from is now — only shows new events going forward
Use --from 5m or --from 1h to include recent historical events
Probe status errors (e.g., instrumentation failures) go to stderr
Exit code 0 if events received, 1 if timed out with no events
Use --wait <seconds> to retry probe existence check (handles create → watch pipeline); default is 0 (fail immediately if not found)

Using `--fields` for compact output

For most agent use cases, --fields gives you exactly what you need without jq:

# Message + captures + timestamp (most common)
pup debugger probes watch <ID> --fields "message,captures,timestamp" --limit 5

# Template message only (one line per event)
pup debugger probes watch <ID> --fields "message" --limit 10

# Just captures (expression values or snapshot data)
pup debugger probes watch <ID> --fields "captures" --limit 1

Extracting fields with jq (reference)

When using the default output (no --fields), the debugger payload is at the top level. Common jq patterns:

# Extract a specific captured expression
pup debugger probes watch <ID> --limit 1 \
  | jq '.snapshot.captures.return.captureExpressions'

# Get captured argument values
pup debugger probes watch <ID> --limit 1 \
  | jq '.snapshot.captures.return.arguments'

Default output structure (trimmed to debugger payload):

.snapshot.captures.return.arguments          → method arguments
.snapshot.captures.return.locals             → local variables
.snapshot.captures.return.captureExpressions → capture expression values
.snapshot.probe.id                           → probe ID
.snapshot.probe.location                     → source location

Supported Languages

Language	`--language` value
Java	`java`
Python	`python`
.NET	`dotnet`

Failure Handling

Problem	Fix
Wrong env / no instances	Use `pup debugger context <service>` to list valid environments
"probe not found"	Use `--wait <seconds>` to retry, e.g. `pup debugger probes watch <ID> --wait 10`
No events appearing	Check `--from` (default is `now`); probe may need time to instrument
Instrumentation errors	Check stderr output from watch for status errors
Auth error	Run `pup auth login` or set `DD_API_KEY` + `DD_APP_KEY` + `DD_SITE`
Wrong method signature	Use the `dd-symdb` skill to find exact `TYPE:METHOD` or `TYPE:METHOD(args)` values

References

Copy & paste — that's it

npx skills add https://github.com/datadog-labs/pup --skill dd-debugger

Run this in your project — your agent picks the skill up automatically.

Prerequisites

pup must be installed:

brew tap datadog-labs/pack && brew install pup