open-context-check

Live context-window usage analyzer for OpenCode. Reads OpenCode's own token telemetry from opencode export, then prints current session usage.

---

/context — compact

Quick live check. Fits in a glance.

  ▓▓▓▓▓▓▓▓▓▓▓▓▓▓░░░░░░░░░░░░░░░░  Context Usage (live)
                                  93.6k/200.0k tokens (46.8%)
                                  from opencode export latest API call

  ░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░  Fresh input          849 (0.4%)
  ▓▓▓▓▓▓▓▓▓▓▓▓▓▓░░░░░░░░░░░░░░░░  Cache read         92.7k (46.3%)
  ░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░  Cache write            0 (0.0%)
  ░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░  Output                50 (0.0%)
  ░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░  Reasoning              0 (0.0%)
  ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓░░░░░░░░░░░░░░  Free space        106.4k (53.2%)

/context standard / /context detailed

Live usage plus session metadata.

  ▓▓▓▓▓▓▓▓▓▓▓▓▓▓░░░░░░░░░░░░░░░░  Context Usage (live)
                                  93.6k/200.0k tokens (46.8%)
                                  from opencode export latest API call

  Fresh input / cache / output / reasoning rows...

  Session: ses_...
  Model: openai/gpt-5.5 (high)
  Latest call: 2026-05-03T23:04:27.371Z
  Message: msg_...
  Completed API calls: 10

detailed is kept as a familiar alias for metadata output. Static estimates were removed because they are not live telemetry.

---

Why?

OpenCode records token telemetry per API call, but does not expose it as a compact /context view. This tool prints that live telemetry without guessing at prompt/tool overhead.

Install

git clone https://github.com/Sharper-Flow/open-context-check.git
cd open-context-check
npm install
npm run build

As an OpenCode /context command

Copy the command file to your global commands directory:

cp commands/context.md ~/.config/opencode/commands/

The bundled command defaults to $HOME/dev/oc-plugins/open-context-check. If you cloned elsewhere, set OPEN_CONTEXT_CHECK to your built script path or edit the command file.

Then use /context inside OpenCode:

/context              # compact live usage
/context standard     # live usage with session details
/context detailed     # live usage with session details

Standalone CLI

Run directly from anywhere:

node /path/to/open-context-check/dist/context-check.js              # compact
node /path/to/open-context-check/dist/context-check.js standard     # standard
node /path/to/open-context-check/dist/context-check.js detailed     # detailed
node /path/to/open-context-check/dist/context-check.js --limit 128k # override context window size

What it analyzes

Component	Source
Live context usage	opencode export <session> --sanitize latest completed assistant token telemetry
Session ID	$OPENCODE_SESSION_ID, explicit --session, or newest opencode session list entry
Context limit	Configured model context limit when present, otherwise known model-family defaults

Known limitations

• Latest completed API call only. In-flight assistant responses appear after completion.
• Provider telemetry split only. Rows show fresh input/cache/output/reasoning, not system-vs-tools-vs-instructions.
• No static estimates. The tool avoids guessed MCP/agent/instruction overhead.

Contributing

PRs welcome. Codebase is small:

src/
├── context-check.ts   # CLI entry point
├── config.ts          # Config discovery for context-limit detection
├── report.ts          # Live report formatting
├── session.ts         # Live session telemetry from opencode export
└── tokens.ts          # Formatting helpers + tokenizer cleanup
commands/
└── context.md         # OpenCode /context custom command

License

MIT