Agent Runs

How the SpecSource agent works and what it produces.

An agent run is a single execution of SpecSource's AI. It reads a Linear issue, gathers context from your connected tools, and writes a detailed specification.

What triggers a run

  • Manual — Click "Run Agent" in the dashboard
  • Schedule — Runs on a recurring interval

What happens during a run

  1. Read the issue — Extracts the issue description, labels, and context from Linear
  2. Check error data — Searches Sentry for related errors, stack traces, and event history
  3. Gather code context — Searches GitHub for relevant source files, recent commits, and PRs
  4. Search for team context — Finds related Slack discussions and prior investigations
  5. Check for duplicates — Compares against existing Linear issues using embedding-based semantic matching
  6. Write the specification — Synthesizes all context into a structured specification
  7. Enrich the Linear issue — Updates the issue with the detailed specification

Each step is logged and visible in the run detail view.

Run statuses

StatusMeaning
PendingQueued, waiting to execute
RunningCurrently gathering context and writing the specification
CompletedFinished successfully
FailedSomething went wrong (check the error message)
UnpaidCredits exhausted — the run was not executed

Viewing results

Click any run in the Agent Runs page to see:

  • Each step the agent took
  • What data it found
  • The specification it wrote (with a link to Linear)
  • Token usage and duration