session-logs

name	session-logs
description	Search and analyze past session logs. Use when the user says things like "in the previous conversation..." or "I did something before...", or when investigating cost and tool usage.
enabled	true

session-logs

A skill for searching and analyzing GeminiClaw session history. Operates using jq and rg (ripgrep) via run_shell_command.

File Structure

Sessions are stored in JSONL format, with on-demand LLM-generated summaries:

JSONL (structured data)

Location: {workspace}/memory/sessions/<sessionId>.jsonl

One entry per line, structure:

{
  "runId": "uuid",
  "timestamp": "2026-02-23T12:34:56.789Z",
  "trigger": "manual" | "heartbeat",
  "responseText": "Agent response text",
  "toolCalls": [
    {"name": "tool_name", "args": {...}, "result": "...", "status": "success" | "error"}
  ],
  "heartbeatOk": false,
  "tokens": {"total": 12345, "input": 12000, "output": 345}
}

Summaries (LLM-generated, Obsidian-optimized)

Location: {workspace}/memory/summaries/YYYY-MM-DD-<slug>.md

Auto-generated when a session goes idle, or manually via geminiclaw session summary <sessionId>. Contains YAML frontmatter (date, tags, tokens), TL;DR, key decisions, and full conversation log.

Common Queries

List sessions (sorted by date)

ls -lt {workspace}/memory/sessions/*.jsonl | head -20

Search sessions for a specific date

ls {workspace}/memory/sessions/*.jsonl | xargs grep -l "2026-02-23"

Keyword search in response text

jq -r '.responseText' {workspace}/memory/sessions/*.jsonl | rg -i "keyword"

Find sessions where a specific tool was called

jq -r 'select(.toolCalls[].name == "browser_navigate") | .timestamp + " " + .runId' {workspace}/memory/sessions/*.jsonl

Total token usage across sessions

jq -s '[.[].tokens.total] | add' {workspace}/memory/sessions/*.jsonl

Daily token usage

jq -r '"" + .timestamp[:10] + " " + (.tokens.total | tostring)' {workspace}/memory/sessions/*.jsonl \
  | awk '{a[$1]+=$2} END {for(d in a) print d, a[d]}' | sort -r

Tool usage frequency ranking

jq -r '.toolCalls[].name' {workspace}/memory/sessions/*.jsonl | sort | uniq -c | sort -rn | head -20

Sessions with the most errors

jq '{runId, timestamp, errors: [.toolCalls[] | select(.status == "error")] | length}' \
  {workspace}/memory/sessions/*.jsonl | jq -s 'sort_by(-.errors) | .[:10][]'

Summary of the last N sessions

ls -t {workspace}/memory/sessions/*.jsonl | head -5 | xargs -I{} sh -c \
  'echo "=== {} ===" && jq "{timestamp, trigger, tokens, tools: [.toolCalls[].name]}" {}'

Searching Summaries

# Search all summaries by keyword
rg -i "keyword" {workspace}/memory/summaries/

# List summaries for a specific date
ls {workspace}/memory/summaries/2026-02-23-*.md

# Search by tag in frontmatter
rg "topic/memory" {workspace}/memory/summaries/

Finding the Workspace Path

{workspace} can be found via the geminiclaw_status tool or in the "Workspace" section of GEMINI.md.

Tips

• The first line of a JSONL file may be null — filter with select(. != null) in jq
• Summaries are auto-generated on session idle; use geminiclaw session summary <id> for manual generation
• Summaries are Obsidian-compatible with YAML frontmatter and tags