Sessions

Sessions

A session represents a single Claude Code conversation. Sessions are automatically created on first hook receipt and updated as hooks arrive.

Session fields

Most fields are computed at query time from the underlying hook rows, not persisted separately:

Field	Type	Description
id	ID!	Claude Code session ID
cwd	String	Working directory of the session
firstSeenAt	Time!	Timestamp of the first hook event
lastSeenAt	Time!	Timestamp of the most recent hook event
durationSeconds	Float	Seconds between first and last hook
hookCount	Int!	Total hooks recorded in this session
model	String	Claude model used (e.g. claude-sonnet-4-6)
gitBranch	String	Git branch active during the session
toolUsage	[ToolStat!]!	Per-tool call counts, sorted by frequency
skillsUsed	[SkillUsage!]!	Skills invoked via the Skill tool
errorCount	Int!	Number of detected tool errors
errors	[ToolError!]!	Detailed error events
tokenUsage	TokenUsage	Cumulative token counts from transcript
summary	String	Cached LLM-generated summary
hooks	[Hook!]!	Raw hook events (supports same filters as top-level hooks)
transcript	[TranscriptMessage!]!	Structured .jsonl transcript messages
subagents	[Subagent!]!	Subagent sessions spawned during this session
prompts	[String!]!	User prompts submitted during the session
editedFiles	[String!]!	File paths touched by Edit/Write tool calls

Session intelligence fields

The most useful high-level fields for understanding what happened in a session are:

{
  sessions(limit: 5) {
    id
    status
    cwd
    model
    gitBranch
    lastSeenAt
    toolUsage { name count }
    skillsUsed { name count }
    subagents { agentType description }
    editedFiles
    prompts
    errorCount
  }
}

These fields let you reconstruct the work: what the user asked, what agents were delegated, which skills/tools were used, and which files changed.

Token usage

Token counts are read from the session's .jsonl transcript file, not from hooks. Fields:

tokenUsage {
  inputTokens
  outputTokens
  cacheReadTokens
  cacheCreationTokens
}

Session summaries

The summarizeSession mutation builds a text digest (prompts, tools, files touched, errors) and pipes it to an LLM CLI tool for summarization. The result is cached in sessions.summary.

mutation {
  summarizeSession(sessionId: "abc-123") {
    id
    summary
  }
}

Querying sessions

List sessions most-recently-active first:

{
  sessions(limit: 10, offset: 0) {
    id
    cwd
    model
    gitBranch
    hookCount
    durationSeconds
    toolUsage { name count }
    skillsUsed { name count }
    errorCount
    tokenUsage { inputTokens outputTokens cacheReadTokens cacheCreationTokens }
  }
}

Get a single session by ID:

{
  session(id: "abc-123") {
    id
    cwd
    summary
    errors { toolName errorMessage input recordedAt }
  }
}