Skip to main content

AI issue-resolution skills

Log Lens ships two agent skills that let an AI coding assistant investigate, fix, and close out your issues by talking to Log Lens over its JSON API. One skill targets Claude Code; the other targets Codex in VS Code. Both do the same job: Log Lens is the evidence source and workflow ledger, while your open project folder is the code they inspect, edit, and test.

Why it matters​

Instead of copy-pasting stack traces into a chat, you can say "fix the top 3 issues by occurrence" or "resolve the open Billing errors," and the assistant works the whole batch: it discovers the issues, reads the raw evidence, maps production paths to your local code, applies a fix, validates it, and writes the workflow status back to Log Lens.

How to use it​

The skills are installed with Log Lens. In your assistant, just describe the work in plain language:

  • "Fix issue 15."
  • "Work through everything unresolved in the checkout app."
  • "Resolve all issues from 21 July 2026, highest occurrence first."
  • "Fix the top N errors, then mark them fixed."

The assistant selects exactly one application (there is no all-apps mode), builds a queue from the lean errors endpoint, and processes issues one at a time.

What happens under the hood​

StepBehavior
OriginReads LOG_LENS_URL from the Log Lens config or env. It is empty by default - the skills run outside a request, so this is the one setting they need you to set.
AuthSends X-Log-Lens-Token when an API key is configured; 401 means resolve or ask.
DiscoverPages through ?api=errors with your filters (status, severity, module, date, tag).
InvestigateReads ?api=error and, when needed, the exact raw event via ?api=source.
Fix & validateDiagnoses, makes the smallest complete fix with test coverage, and runs it.
Close outPosts in_progress, then fixed with a Cause / Fix / Validated note.

Important details​

  • Privacy: all log content stays local. The skill never sends stacks, context, tokens, or payloads to external services.
  • No fix from a diff alone: an issue is only marked fixed after validation runs; otherwise it stays in_progress with a blocker note.
  • Reoccurrence: a later matching import flips a fixed issue to reoccurred automatically.
  • Count meaning: count is lifetime frequency, not a single day's - see issues and fingerprints.
  • Connectors: the skill only syncs logs when you explicitly ask.