Claude Session Introspect

What it does

When your agent uses it

• The user asks how many tokens did this session use
• How many prompts have I sent
• Show me the stats for this conversation
• What got compacted

Example prompts

• “how many tokens did this session use”
• “how many prompts have I sent”
• “show me the stats for this conversation”
• “/claude-session-introspect”

Requirements

• A Bash shell
• Compatibility (from SKILL.md): Requires `jq`. Sessions live at `~/.claude/projects/<encoded-cwd>/<session-uuid>.jsonl`. The encoded-cwd is the absolute working directory with `/` replaced by `-` and a leading `-`. Each line is a JSON object with `type`, `message`, `toolUseResult`, etc.

What it can do on your machine

Read from SKILL.md and the folder at commit 038ef34. It shows what the files ask for, not the result of running them.

• Tool permissions

  Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

  From allowed-tools in the SKILL.md frontmatter.

• Runs code

  Ships script files (Shell), which the agent can run.

  Shell commands in SKILL.md call:

  • jq
  • bash
  • claude

  From the folder's file list and the shell code blocks in SKILL.md.

• Network

  Links to these hosts (documentation or services it may open):

  • talraviv.co

  From URLs in SKILL.md, links to its own repository left out.

• Credentials

  Names no API keys, tokens, secrets or passwords.

  From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

• Compatibility

  Requires `jq`. Sessions live at `~/.claude/projects/<encoded-cwd>/<session-uuid>.jsonl`. The encoded-cwd is the absolute working directory with `/` replaced by `-` and a leading `-`. Each line is a JSON object with `type`, `message`, `toolUseResult`, etc.

  From compatibility in the SKILL.md frontmatter.

Context cost

Claude Session Introspect loads about 2k tokens when it runs. Until then it costs about 188 tokens; SKILL.md has 592 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it

~188

When it runs · the whole SKILL.md, loaded when a task matches

~2k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from swyxio/skills at commit 038ef34, republished under its MIT licence (© swyxio). 592 words, ~2,045 tokens.

name

claude-session-introspect

description

Inspect Claude Code session JSONL files at ~/.claude/projects/ to extract real conversation telemetry: token counts (input/output/cache reads/cache writes), assistant turn counts, human prompt counts, tool-use counts, compaction boundaries, and the contents of compaction summaries. Use this skill when the user asks "how many tokens did this session use", "how many prompts have I sent", "show me the stats for this conversation", "what got compacted", "where are the compaction boundaries", "introspect the session", "do brain surgery on the JSONL", or wants any data point that lives inside the on-disk session log rather than the live context window. Inspired by Tal Raviv's "I wanted to know how compaction works" article.

compatibility

Requires `jq`. Sessions live at `~/.claude/projects/<encoded-cwd>/<session-uuid>.jsonl`. The encoded-cwd is the absolute working directory with `/` replaced by `-` and a leading `-`. Each line is a JSON object with `type`, `message`, `toolUseResult`, etc.

license

MIT

metadata.author

swyxio

metadata.version

1.0

metadata.last-updated

2026-04-08

metadata.primary-tools

jq, bash

~/.claude/projects/<encoded-cwd>/<session-uuid>.jsonl

# 1. encode current working directory
ENC="-$(pwd | sed 's,/,-,g' | sed 's/^-//')"
# 2. list the project's session files, newest first
ls -t "$HOME/.claude/projects/$ENC/"
# 3. the most recent .jsonl is usually the live one
SESSION="$HOME/.claude/projects/$ENC/$(ls -t "$HOME/.claude/projects/$ENC/" | head -1)"
echo "$SESSION"

find ~/.claude/projects -name '<uuid>.jsonl'

bash stats.sh "$SESSION"

jq -s '
  [.[] | select(.message.usage)] |
  {
    assistant_turns: length,
    input_tokens:        (map(.message.usage.input_tokens // 0)              | add),
    output_tokens:       (map(.message.usage.output_tokens // 0)             | add),
    cache_read_tokens:   (map(.message.usage.cache_read_input_tokens // 0)   | add),
    cache_create_tokens: (map(.message.usage.cache_creation_input_tokens // 0)| add)
  }
' "$SESSION"

jq -r '.type' "$SESSION" | sort | uniq -c

jq -r '
  select(.type == "user" and .toolUseResult == null) |
  (.message.content
    | if type == "string" then .
      else (map(select(.type == "text") | .text) | join("\n"))
      end)
' "$SESSION" > /tmp/prompts.txt

# total non-empty user message blocks
grep -cv '^$' /tmp/prompts.txt

# distinct human messages = blocks not starting with <system-reminder> or <command-
awk '
  BEGIN { n = 0; cur = "" }
  /^$/ { if (cur != "" && cur !~ /^<system-reminder>/ && cur !~ /^<command-/) n++; cur=""; next }
  { if (cur=="") cur=$0 }
  END { if (cur != "" && cur !~ /^<system-reminder>/ && cur !~ /^<command-/) n++; print n }
' /tmp/prompts.txt

jq -r '
  select(.type == "assistant") |
  .message.content[]? |
  select(.type == "tool_use") |
  .name
' "$SESSION" | sort | uniq -c | sort -rn

# count compaction events
jq -r 'select(.type=="system" and (.subtype // "") == "compact_boundary") | .timestamp' "$SESSION" | wc -l

# was each one auto or manual?
jq -r '
  select(.type == "system" and (.subtype // "") == "compact_boundary") |
  {ts: .timestamp, trigger: (.compactMetadata.trigger // "unknown"), preTokens: (.compactMetadata.preCompactTokens // null)}
' "$SESSION"

# read the compaction summaries (the actual contents that survived)
jq -r '
  select(.type == "user" and (.isCompactSummary == true or
    ((.message.content // "") | tostring | test("session is being continued from a previous conversation"))))
  | (.message.content | if type == "string" then . else (map(select(.type=="text").text)|join("\n")) end)
' "$SESSION" | less

jq -r '
  select(.message.usage) |
  [.timestamp,
   (.message.usage.input_tokens // 0),
   (.message.usage.output_tokens // 0),
   (.message.usage.cache_read_input_tokens // 0)]
  | @tsv
' "$SESSION" | column -t

© swyxio, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 2 other files in claude-session-introspect of swyxio/skills.

• SKILL.md
• README.md
• stats.sh

Open the folder on GitHubat commit 038ef34