session-retrospect

wimcc — What's in My Claude Code

English · 한국어

Replay what Claude Code actually did — not what it said.

What's in my cc?

wimcc replays a Claude Code session — the one that just ran, or the one running now — as an execution you can step through. It surfaces what the chat log alone doesn't: which tool failed and why, how much time and how many tokens a model request cost, what a hook blocked, which edit changed which lines of which file — all on one screen, each item traceable straight back to the raw record it came from.

Things you can do with wimcc:

• See which tool calls failed in this session, and why.
• Track where the tokens went — usage, cost, and context efficiency.
• Trace which edit changed which lines of which file (file lineage).
• Check what a hook blocked or let through.
• Follow how model requests, tool calls, and hooks interleaved over time.
• Read all of the above in a browser UI — or pull it into another tool or agent over the Pull API / MCP.

Everything runs locally on 127.0.0.1 and nothing is sent outward. External access is read-only.

Quick start

Prebuilt binary

Grab the archive for your platform from GitHub Releases — Linux x86_64 or macOS Apple Silicon. The WebUI is embedded; the single binary is all you need.

tar -xzf wimcc-v*-aarch64-apple-darwin.tar.gz   # or …-x86_64-unknown-linux-gnu.tar.gz
./wimcc init-db                # apply migrations, prepare .wimcc.sqlite
./wimcc serve --auto-migrate   # http://127.0.0.1:7878  (auth off by default)
./wimcc doctor                 # verify collector wiring

Build from source

just build-release                            # build the SPA + release binary (target/release/wimcc) in one step
./target/release/wimcc init-db                # apply migrations, prepare .wimcc.sqlite
./target/release/wimcc serve --auto-migrate   # http://127.0.0.1:7878  (auth off by default)
./target/release/wimcc doctor                 # verify collector wiring

serve runs everything in one process: the read-only Pull API, the embedded WebUI, the OTel + hook receivers, and a live tail of ~/.claude/projects transcripts. For Claude Code to emit OTel + hook events into wimcc you set up ~/.claude/settings.json once — see Connecting Claude Code. wimcc doctor tells you what's connected and what's missing.

wimcc ingest --all is still available for backfill (a cold-start sweep of existing transcript JSONLs) but is not required for live operation.

CLI

wimcc [--db-path <PATH>] [--log-format pretty|json] [--verbose] <command>

Global flags apply to every subcommand. --db-path defaults to .wimcc.sqlite (env WIMCC_DB).

Command	Purpose
init-db	Apply migrations and prepare the database.
ingest --all / ingest --path <P>	Backfill: scan transcript JSONL files into raw + observed events (idempotent).
doctor [--json] [--server <URL>] [--project <DIR>]	Read-only diagnosis of collector wiring (settings hierarchy, hooks, server probe). Never mutates anything.
serve	Start the local service: Pull API + WebUI + OTel/hook receivers + transcript live tail.

wimcc serve flags

wimcc serve [--bind 127.0.0.1] [--port 7878]
             [--auto-migrate]                    # apply pending migrations on startup
             [--transcripts-root <PATH>]         # override ~/.claude/projects
             [--no-watch-transcripts]            # disable the live tail (OTel/hook only)
             [--auth off|on]                      # bearer-token auth on /v1 + /mcp (default: off)
             [--retention-profile none|default|strict]   # background retention sweep (default: none)
             [--print-token] [--rotate-token]     # manage the bearer token, then exit
             [--sse-keepalive-secs N]             # WebUI live-stream keep-alive (default: 30)
             [--sse-channel-capacity N]           # broadcast channel capacity (default: 512)
             [--shutdown-after-ms N]              # test/smoke convenience

What it observes