cc-what-have-i-done

README

ccwhid — cc-what-have-i-done

Turn a Claude Code session transcript into a browsable, self-contained static HTML report. See what you prompted and what Claude Code did — and commit the report into your repo as a documentation artifact.

Status: under active development. See docs/superpowers/plans/ for the implementation plan.

Preview

A full example — ccwhid rendering the very session that built it — is committed at docs/session-report/index.html; clone the repo and open it in a browser.

Report overview	Tool calls & diffs

The report is one self-contained page: a jump-to-prompt sidebar, a live turn filter, expand/collapse-all, and a light/dark toggle — all client-side, no server.

Install

go install github.com/saigyo/cc-what-have-i-done/cmd/ccwhid@latest

Or build locally:

go build -o ccwhid ./cmd/ccwhid

Usage

ccwhid                      # browse sessions in an interactive TUI
ccwhid --latest             # render the most recent session
ccwhid --session <id>       # render a specific session (id or prefix)
ccwhid --session <id> --open

The report is written to ./ccwhid-report/<session-short>/ by default (override with --out). Open index.html in any browser — no server needed.

Browsing sessions

Run ccwhid with no selector to open the interactive browser:

• Project list — every project with sessions, most-recent first. ↑/↓ to move, enter (or →) to open a project, q to quit.
• Session list — the chosen project's sessions. enter selects, ← (or esc) returns to the project list.
• Both lists scroll to fit any terminal size.
• By default only your interactive sessions are shown. Claude Code writes each Task subagent / code-review agent as its own transcript; press a in the session list to show (and hide) those agent sessions — they're marked with a ⟲.
• Selecting a session opens an options screen to toggle subagents, redaction, and open-in-browser, and to type an output directory (blank uses the default) before generating.

--project <name> opens the browser directly on that project's session list.

Pick a project, then a session. Agent transcripts are hidden until you press a, then marked with ⟲:

Flags

Flag	Description
--session <id>	Session id or unambiguous prefix to render
--project <path|name>	Scope --latest to a project, or open the TUI on it (matches full path, basename, or unambiguous substring)
--latest	Render the most recent interactive session (skips agent transcripts)
--out <dir>	Output directory
--title <str>	Override the report title
--include-subagents	Include subagent work: inline Task sidechains and linked agent-session transcript pages under subagents/ (default true; --include-subagents=false to omit)
--no-redact	Disable all redaction (secrets, paths, and names)
--redact-name <str>	Display name to scrub from output (default: your OS/git display name)
--no-redact-name	Disable display-name redaction (account/path redaction still applies)
--force	Overwrite a non-empty output directory
--open	Open the report in a browser when done
--usage	Include a token-usage & estimated-cost section (default off)

Redaction

By default, ccwhid scrubs common secret shapes (AWS keys, API tokens, JWTs, KEY=/TOKEN=/SECRET= assignments) and rewrites your home directory to ~. It also scrubs account names from home-directory paths in any form — plain (/Users/alice, /home/alice), Windows (C:\Users\alice), and Claude Code's dash-encoded project dirs (-Users-alice-IdeaProjects-…) — plus standalone mentions of your own login name (e.g. the owner column of ls -l output), replacing them with [user].

It also scrubs your display name (resolved from --redact-name if given, otherwise your OS account name, otherwise git config user.name) in the forms that leak into module paths, usernames, and commit trailers — the verbatim name plus its concatenated and separator-joined variants (Jane Doe, janedoe, jane-doe, jane.doe, jane_doe), so github.com/janedoe/project becomes github.com/[user]/project. Only multi-word names are matched, and only the known name — generic GitHub usernames are left intact. Turn this off with --no-redact-name.

This is best-effort defense-in-depth — review generated reports before committing them. Disable all redaction with --no-redact.

What's included

Your prompts, Claude's replies (rendered markdown + syntax-highlighted code), tool calls with collapsible detail, Edit/Write diffs, collapsed thinking blocks, and — optionally — nested subagent activity. Background-agent results (e.g. a Task finishing) appear as their own "Agent" cards; linked subagent sessions — found under the transcript's <sessionId>/subagents/ directory — render as separate pages under subagents/, linked from both the Agent tool card and the result card. With --usage, their cost is rolled into the totals and shown as an "of which subagents" line. System reminders and attachments are omitted for readability.

Known limitation: SDK-spawned sessions without parent linkage (e.g. some review-hook runs) aren't attributed to a parent turn and so aren't included.

Token usage & cost

Pass --usage (or tick "Include usage & cost" in the TUI) to add a collapsible Usage card and per-turn cost badges. Token counts come straight from the transcript's usage data and are exact; cost is an estimate from a built-in Anthropic list-price table (dated in the report footnote) — unknown models show tokens with cost n/a, and server-tool fees are not included. Prices are embedded, so this works fully offline.

These figures cover the rendered transcript plus any linked subagent sessions (shown separately as "of which subagents"). SDK-spawned sessions without parent linkage aren't attributed or included, which is why the total can still differ from Claude Code's /usage session rollup.

License

MIT © 2026 Markus Ackermann