Index of every page on this site: llms.txt
All skills
launchdarkly avatar

/investigate

@f0ee110 official
Run it yourself

Finds the root cause of a production problem from observability data (logs, traces, errors, sessions, metrics): scope and onset, what changed (deploys and flag changes), which host, version, route or flag variation is over-represented against a baseline, then trace, log and error evidence and blast radius. Use when the user reports a bug, an outage, an error or latency spike, an incident, or asks about patterns across application data.

Use this Skill: https://skilld.dev/gh/launchdarkly/ai-tooling/investigate

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

errors.md

≈1.1k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Investigating with error groups

Load this file when the investigation concerns exceptions, crashes, stack traces, or a new error.

When to reach for error groups

  • The user reports a crash, an exception or a regression
  • You need an error's first-seen time, frequency, stack, or affected users
  • You have an error group ID and want its occurrences, traces and sessions

An error group collapses many occurrences by stack similarity. A group can still hold unrelated messages, so read its occurrences before concluding.

Find groups (search-error-groups)

  • startDate is required. Only OPEN groups are returned unless state says otherwise (ANY for every state).
  • Filter on exception.message, exception.type, service_name and environment. Do not filter on event: it matches only the group's original title.
    query: exception.message="Cannot use 'in' operator*" AND service_name=gonfalon-web
  • sort: spike for groups growing fastest (count in the window divided by the count in the previous window of the same length; in_window adds baseline_count, spike_ratio and new_in_window, and the response gives spike_baseline_window), count for the biggest in the window, first_seen for new groups, sessions for user impact (backend-only groups have no sessions and are left out).
  • Each group's in_window block counts only the matching occurrences in your window.

Read one group (get-error-group)

  • secureId from search-error-groups. It returns the state, the stack and one occurrence (the latest, or the latest in startDate..endDate that matches query) with its trace_id and session.
  • include: ["distribution", "trend"] adds context over startDate..endDate (default the last 7 days, echoed as context_window): distribution gives the top values of browser, os_name, environment, service_name and service_version with count and percent, so "92 % on version X" or "only Safari" shows up in one call; trend gives occurrence counts in up to 60 buckets with the peak bucket; the first non-zero bucket is the group's onset in that window. Not with occurrences.
  • Frames carry in_app, link (to the source on GitHub when connected) and source_map_error; top_frame adds code_before and code_after.
  • Step through occurrences with next_id and previous_id, or set occurrences (with startDate) to list them: each row has timestamp, service, version and session.
  • Pivot: _pivots holds ready get-trace and search-logs (trace_id=<id>, ±1 h) arguments for the occurrence; pass them as they are. Without _pivots, build them from the occurrence's trace_id.
  • search-logs rows at error level and above carry error_group_secure_id and error_object_id when the log matches an error object, which links a log line straight to its group.

Typical patterns

  1. New regression: search-error-groups with sort: "spike" (or "first_seen") over the onset window, then get-changes for the service: did a version appear just before the first occurrence?
  2. What is over-represented: get-error-group include: ["distribution"] for one group, or search-aggregations with productType: "errors", grouped by service_name, the version key or host.name, with compareTo: "1d". For a flag, filter feature_flag.key=<flag> and group by feature_flag.result.value.
  3. Blast radius: sort: "sessions", or search-aggregations CountDistinct of secure_session_id for the group's filter with exact: true.
  4. Stack analysis: the originating frame is the deepest frame in application code, not framework code. If it is in a third-party dependency, say so.

Interpreting results

  • Scope first, cause second. One user in a rare path is different from 30 % of traffic.
  • Onset vs standing. Compare the group's count with compareTo; a group that was just as busy last week is not this incident's cause.
  • Cite the error group secure ID, the occurrence count, first-seen time, and one trace ID.

Common mistakes

  • Filtering on event. Use exception.message.
  • Reporting "many errors" without a number and a comparison window.
  • Treating backend errors that are only logged (not recorded as error groups or error spans) as absent. Check error logs too.

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

1 warning2d3 checks · Risk SAFE
  • Gen Agent Trust Hub2d

    This skill provides specialized instructions for investigating production outages and performance issues using the LaunchDarkly observability platform. It provides a structured methodology for identifying root causes through log analysis, trace inspection, and error grouping, utilizing the official LaunchDarkly MCP server.

  • Socket2d

    No alerts

  • Snyk2d

    Risk: MEDIUM · 1 issue

Signed by skilld at f0ee110. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated 2 days ago
compatibility
Requires the remotely hosted LaunchDarkly MCP server
metadata
{
  "author": "launchdarkly",
  "version": "0.5.0"
}

README badge

README badge for launchdarkly/ai-tooling/investigate