SuperAPI AI Directory

jsonl-core

General & other

View repository

Updated 10-16 07:18

About this skill

JSONL Core — Conversation Parsing Infrastructure

Architecture

All parsing logic lives in ${CLAUDE_PLUGIN_ROOT}/scripts/echolib.py — a single Python module (stdlib only, Python 3.6+). The shell scripts are thin wrappers around it.

Data Locations

  • Session index (fast path): ~/.claude/projects/<encoded-path>/sessions-index.json
  • Fallback index (built by echo-sleuth): ~/.claude/projects/<encoded-path>/.echo-sleuth-index.json
  • Full conversations: ~/.claude/projects/<encoded-path>/<uuid>.jsonl
  • Subagent conversations: ~/.claude/projects/<encoded-path>/<uuid>/subagents/agent-<id>.jsonl
  • Global prompt history: ~/.claude/history.jsonl

The <encoded-path> is the project's absolute path with / replaced by - (e.g., -Users-joker-github-myproject).

Important: Only ~10% of projects have sessions-index.json. The scripts automatically build a fallback index from raw .jsonl files for the remaining 90%, cached in .echo-sleuth-index.json.

Strategy: Fast Path First

Always start with the index before opening any .jsonl file:

bash ${CLAUDE_PLUGIN_ROOT}/scripts/list-sessions.sh "current" --limit 20
bash ${CLAUDE_PLUGIN_ROOT}/scripts/list-sessions.sh "all" --grep "search term" --limit 10
bash ${CLAUDE_PLUGIN_ROOT}/scripts/list-sessions.sh "/path/to/project"

Output is tab-separated: SESSION_ID CREATED MODIFIED MSG_COUNT BRANCH SUMMARY FIRST_PROMPT PROJECT_PATH FULL_PATH

The FULL_PATH field (9th column) is the absolute path to the .jsonl file. Use this to pass to other scripts.

Only open the full .jsonl when you need message-level detail.

Canonical Parser

bash ${CLAUDE_PLUGIN_ROOT}/scripts/parse-jsonl.sh <file.jsonl> [options]

Key modes:

  • Schema detection (check if format has changed):
    bash ${CLAUDE_PLUGIN_ROOT}/scripts/parse-jsonl.sh <file.jsonl> --detect-schema
    
  • Filtered extraction (skip noise, ~38% faster on large files):
    bash ${CLAUDE_PLUGIN_ROOT}/scripts/parse-jsonl.sh <file.jsonl> --types user,assistant --skip-noise --limit 20
    
  • Field selection (only extract specific fields):
    bash ${CLAUDE_PLUGIN_ROOT}/scripts/parse-jsonl.sh <file.jsonl> --types user --fields timestamp,message --format tsv
    

Convenience Scripts

All scripts are at ${CLAUDE_PLUGIN_ROOT}/scripts/. They require only bash + python3 (stdlib only, no pip packages, minimum Python 3.6+). The git scripts additionally use git.

Extract human-readable messages

bash ${CLAUDE_PLUGIN_ROOT}/scripts/extract-messages.sh <file.jsonl> [--role user|assistant|both] [--no-tools] [--limit N] [--thinking [LIMIT]]

Note: --thinking without a number shows full thinking blocks. --thinking 500 truncates to 500 chars. Default: thinking blocks are hidden.

Extract tool calls with results

bash ${CLAUDE_PLUGIN_ROOT}/scripts/extract-tools.sh <file.jsonl> [--tool NAME] [--errors-only] [--limit N]

List files edited in a session

bash ${CLAUDE_PLUGIN_ROOT}/scripts/extract-files-changed.sh <file.jsonl> [--with-versions]

Uses reverse-read on large files (>50MB) to find the last snapshot efficiently.

Quick session statistics (single-pass)

bash ${CLAUDE_PLUGIN_ROOT}/scripts/session-stats.sh <file.jsonl>

Build fallback index

bash ${CLAUDE_PLUGIN_ROOT}/scripts/build-index.sh [project-path|"all"]

Pre-warm the cache for projects without sessions-index.json.

Subagent Discovery

Sessions with subagent work have a <session-uuid>/subagents/ directory. Check for it:

ls "$(dirname <full_path>)/$(basename <full_path> .jsonl)/subagents/" 2>/dev/null

Subagent files follow the same JSONL format and can be parsed with the same scripts.

Performance Notes

  • Python3 startup (80ms) dominates for files < 1MB (97% of all files)
  • --limit N enables early exit — near-instant for small N
  • --skip-noise avoids json.loads on progress/queue-operation lines by string pre-filter
  • For files > 10MB: json.loads is the CPU bottleneck (63% of time), not I/O
  • extract-files-changed.sh uses reverse-read on files > 50MB
  • session-stats.sh counts errors in the same pass (no double-read)
  • grep is NOT faster than Python for this format — avoid grep-then-parse pipelines

When Scripts Are Not Enough

For targeted searches within large .jsonl files, use Grep directly:

# Find user messages containing a keyword (two-step workflow):
# Step 1: Find lines matching the record type
Grep pattern='"type":"user"' path="<file.jsonl>" output_mode="content"
# Step 2: From those results, visually scan or re-grep for your keyword.
#         Alternatively, combine both conditions in one regex:
Grep pattern='"type":"user".*keyword' path="<file.jsonl>" output_mode="content"

# Find error results
Grep pattern='"is_error"\s*:\s*true'

# Find specific tool usage
Grep pattern='"name"\s*:\s*"ToolName"'

# Find decisions (AskUserQuestion usage)
Grep pattern='"name"\s*:\s*"AskUserQuestion"'

Record Type Quick Reference

See references/record-types.md for the complete schema. The essential types:

Type What It Contains When to Use
user (string content) Human's actual request Understanding intent, finding topics
assistant (text blocks) Claude's responses and reasoning Finding decisions, explanations
assistant (tool_use blocks) Tool invocations Understanding what actions were taken
file-history-snapshot Files edited with version counts Knowing which files were touched
summary AI-generated session title Quick identification (also in sessions-index.json)
system (compact_boundary) Context compaction marker Session was long enough to need compaction

Finding the Right Session Directory

To map a project path to its Claude session directory:

  1. Take the absolute project path (e.g., /Users/joker/github/myproject)
  2. Replace all / with - → Users-joker-github-myproject
  3. Prepend - → -Users-joker-github-myproject
  4. Look in ~/.claude/projects/-Users-joker-github-myproject/

If unsure, use list-sessions.sh which handles the lookup automatically (including fuzzy matching via sessions-index.json originalPath).

Noise Filtering

When reading raw .jsonl, skip these:

  • Records where type is progress or queue-operation (streaming/internal bookkeeping)
  • User records with isMeta: true (slash command injection)
  • User records with isCompactSummary: true (auto-generated context, not human input)
  • Assistant records with model: "<synthetic>" (passthrough, not real inference)
  • User records where content is an array of tool_result blocks (tool outputs, not human messages)

Use --skip-noise with parse-jsonl.sh for automatic noise filtering, or use the convenience scripts which handle this internally.

Schema Evolution Awareness

Claude Code evolves rapidly. The JSONL format has changed across versions:

  • New record types appear silently (e.g., progress at v2.1.14, pr-link later)
  • New optional fields are added to existing records (~3-5 per minor version)
  • Some record types (summary, pr-link, file-history-snapshot) lack common fields like version or uuid
  • The directory encoding is lossy for Unicode paths — use sessions-index.json's originalPath field as ground truth

When parsing results look unexpected, use the schema-scout agent or run:

bash ${CLAUDE_PLUGIN_ROOT}/scripts/parse-jsonl.sh <file.jsonl> --detect-schema

to check for unknown record types or field changes.