Claude Code integration

If you work with Claude Code, Showtail can help build the trail for you. Install the bundled skill, which includes auto-capture hooks by default:

# In your project. Creates ./.claude/skills/showtail/ and ./.claude/settings.json.
showtail connect claude --project

# Or install for all projects.
showtail connect claude --user

# Install the skill without changing settings. Capture is then manual.
showtail connect claude --project --no-hooks

What this gives you:

• A Showtail skill that helps Claude keep the trail tidy and offers to generate the report when you finish.
• Hooks, enabled by default, that capture your prompts and file edits automatically.

When	Showtail does this
You submit a prompt	Logs it as a prompt event
Claude edits or writes a file	Snapshots that file as an artifact
Claude pauses to ask you to choose between options	Logs your choice as a decision
A session starts	Ensures a work session exists

If you install with --no-hooks, the skill can still check showtail status --json and log prompts and file snapshots itself. That mode is model-driven rather than guaranteed, but it still gives you a useful trail.

At the end of your work session, run:

showtail report

The report gives your educator a readable view of what happened.

Install through the Claude Code plugin

You can also install Showtail through the Claude Code plugin marketplace:

/plugin marketplace add Tingsters/Showtail
/plugin install showtail

The plugin includes the same skill and hooks.

Privacy with Claude Code hooks

When hooks are active, every prompt you submit and every file Claude edits is logged locally to .showtail/. That is the purpose of Showtail, but it is worth keeping a few habits in mind:

• Nothing is sent anywhere. There are no external calls and no telemetry.
• The trail may be committed to your repository, so do not put secrets in prompts while capture is on.
• Review your trail anytime with showtail report.
• Turn capture off with showtail disconnect claude. Add --user if you installed at user scope.
• To install the skill without hooks, use showtail connect claude --no-hooks. The skill can still capture prompts itself, but it will not add hooks to settings.json.

Importing an existing session

If you enabled Showtail partway through a session — or only afterwards — you do not have to lose the earlier work. Claude Code writes a full transcript of every session to disk (under ~/.claude/projects/…), and Showtail can back-fill your trail from it so it reads as if capture had been on from the start. This is local like everything else: nothing is fetched, and because the transcript records who said what, there is no guessing about what you wrote versus what Claude produced.

Run it with no arguments to pick from this project's sessions:

showtail import claude

This opens an interactive picker. Each session shows how long ago it ran, its prompt and edit counts, a rough duration, its first and last prompt, and whether it is already in your trail — so the right one is easy to spot:

Claude Code sessions for this project (3):

  1. 2h ago    8 prompt(s), 5 edit(s), ~25 min
     first: rework the claude code import so it is easier to pick sessions
     last:  add tests for the picker
     id: cc010626-49c0-4bef-a72f-2c1a0a4a172e

  2. yesterday    3 prompt(s), 0 edit(s), ~4 min  [imported]
     first: fix the timestamp parsing bug
     id: a019ec8b-df56-4e6b-9176-13bae0c42a86

Pick sessions to import [e.g. 1,3 or 'all', q to quit]:

Pick one or several — 1, a list like 1,3, a range 1-2, or all. The whole selection is imported as a single batch, so one showtail import undo reverses it.

Options:

• <session-id> — import a specific session by its id (or just the start of it), skipping the picker.
• --list — print the same list without prompting, then import by id.
• --no-responses — import only your prompts. Claude's replies are imported by default.
• --file <path> — import a specific transcript .jsonl by path. Useful offline, or for a transcript stored elsewhere.
• -s, --session <id> — import into a specific Showtail session.

The command is also available as showtail import claude-code.

Your prompts become prompt events, Claude's replies become ai_output, and each file Claude edited becomes an artifact — all tagged claude-code and stamped with their original times, so they interleave correctly with the rest of your trail. Re-importing is safe: messages already in your trail are skipped (that is what the [imported] marker means), and you can undo a whole batch in one step:

showtail import undo

In a script or non-interactive shell there is nothing to pick from, so the command imports the most recent session for the project.