@trce/cli
v0.2.2
Published
Local-first coding-agent skill usage, inventory, and maintenance CLI.
Maintainers
Readme
Review system for your agent skills.
trce shows which skills your coding agents use, where copies have drifted, and which skills overlap. Start with a local report. Connect to trce.sh when you want to review skills across your team.
No account for local reports. No prompts, responses, or source code sent to trce.
Get started
Requires Node.js 22.14 or newer.
npm install --global @trce/cli
trce reportOr run without a global install:
npx @trce/cli reportSee recent skill calls, differing copies, and potential cleanup in your terminal. These commands work offline:
trce report --all # Include installed skills with no recorded calls
trce report --since 7d # Focus on the past week
trce report --static # Print at once, without the reveal
trce dedupe # Find skills with overlapping instructions
trce diff review # Compare installed copies of a skill named "review"
trce report --json # Use the report in your own toolstrce reads Claude Code and Codex session history. Cursor support covers installed skills only, not usage. Codex skill reads are inferred evidence, not proof that the instructions were followed.
From one machine to your whole team
Your local report answers "what happens here?" trce.sh brings reports from connected repositories into one workspace so you can:
- See which developers use a skill, with call history across Claude Code and Codex.
- Find versions that have drifted across machines and repositories.
- Review what to share, standardize, or retire, then make repository changes through GitHub pull requests.
- Keep a Shared skills library that teammates can install from and update locally.
Request access to trce.sh to use the team workspace.
With access, link your machine, connect repositories in the app, and preview your first report:
trce init --no-hooks
trce push --dry-run
trce pushThe preview fetches your team's repository scope but uploads no report. To send reports
automatically after sessions, run trce init --hooks-only. No daemon to keep running.
Publishing a report does not publish your skill files. The dashboard receives metadata; reviewed skill files travel directly between your machine and GitHub.
Privacy by design
Local reports stay local. Linked reports contain skill metadata such as names, descriptions, fingerprints, call counts, and token counts, scoped to connected repositories. Linking also sends your machine's short hostname and platform.
trce does not receive prompts, responses, source code, diffs, shell arguments, or unrelated local paths. The package has no telemetry, runtime dependencies, or install scripts.
Read the privacy summary for what is sent and how to stop reporting.
Documentation and help
- Usage guide: linking, installing skills, and finishing reviewed changes.
- Contributing: local setup, tests, and parser fixtures.
- Support · Security · Changelog
Run trce --help for all commands or trce <command> --help for options and network behavior.
Licensed under MIT.
