From a test to useful evidence.

Save the archive and checksum together. Verify before extracting, replacing the placeholder with the actual filename:

# macOS
shasum -a 256 -c <archive>.tar.gz.sha256

# Linux
sha256sum -c <archive>.tar.gz.sha256

Extract it, open a terminal in the extracted folder and install without sudo:

bash scripts/install-cli.sh

Accept optional PATH/completion setup for Bash, Zsh or Fish, then open a new terminal.

worked --version
worked

worked opens the guided menu. If it isn’t on PATH, use ~/.local/bin/worked.

A checksum checks integrity, not the publisher’s identity. If macOS blocks your package, contact SVÄRTA AI; do not disable Gatekeeper.

Confirm a working baseline only after verifying that your environment works.

worked start

# After a change, or when something breaks
worked check
worked report

check captures and compares. report exports the saved comparison as masked Markdown; review it before sharing.

# Once you have verified the environment works again
worked good

worked config init
# Review and edit .worked.json
worked config show
worked config apply
worked test unit

Share .worked.json in your repository for named test commands, collection exclusions, watched paths and ignore rules. init never overwrites a file or runs a test. apply asks you to approve the settings; test separately asks before executing project code.

This focused Node example names your existing npm test command unit and imports its JUnit report. Configure your runner to write that report on every run first:

{
  "version": 1,
  "disabledCollectors": ["config", "disks", "homebrew", "launchd", "network", "npm-global", "packages", "processes", "system", "systemd"],
  "runtimeKeys": ["node", "npm"],
  "watchPaths": ["package.json"],
  "ignore": [],
  "defaultTest": "unit",
  "tests": {
    "unit": {
      "command": ["npm", "test"],
      "report": { "format": "junit", "path": "test-results/junit.xml" }
    }
  }
}

Run worked test for the default suite, or name one explicitly. Reapply edited configuration before running it. Project settings never re-enable collectors you disabled locally. Keep secrets out of this file and only approve trusted repositories. Use --yes for deliberate approval in trusted scripts, not to bypass review.

Worked reads reports produced by your existing test runner; it does not generate them or run browser tests itself. Import a fresh report without a project shortcut:

worked run --name unit \
  --test-report test-results/junit.xml --test-format junit \
  -- npm test

For Playwright, configure its JSON reporter to write test-results/worked.json, then use --test-format playwright with that path, or the equivalent report entry in .worked.json.

Only outcome counts and up to 100 failing names are imported—not error messages, logs, screenshots or attachments. Missing, unchanged or malformed reports are marked unavailable. If requested results are unavailable, contain failures or runner errors, or contain only skipped tests, even an exit-zero command retains the previous reference. Worked still preserves the child’s exit code.

worked api incidents --workspace /absolute/project/path
worked api incident --workspace /absolute/project/path \
  --incident INCIDENT_UUID
worked api incident-activity --workspace /absolute/project/path \
  --incident INCIDENT_UUID

The supplied Node client 0.5 provides client.incidents(), client.incident(id) and client.incidentActivity(id). MCP exposes worked_incidents, worked_incident and worked_incident_activity. Select the worked_analyze_incident prompt with an incident_id in a compatible host. Review import and test execution are not exposed as MCP tools.

These reads use the same masked facts and presentation, require no write permission and remain available after license expiry. Use a full incident ID, never a moving “latest” selector. Inspect worked api capabilities when connecting older installations.

worked run --name unit \
  --baseline-file /private-ci-storage/reference.json \
  --baseline-key myproject.unit.ubuntu24 \
  --report-dir /private-ci-storage/reports \
  -- npm test

Restore the reference before a job and persist it after a successful run. Failed tests or incomplete evidence leave it unchanged. Serialize jobs sharing a file and separate projects and suites. A comparison requires compatible key, command, configuration, collection scope, branch, OS and architecture. An incompatible reference does not prevent your test command from running.

The portable file keeps versions, packages and project-local Git/config evidence, excluding host identity, ports, processes and non-project configuration. It still contains names and hashes: use trusted, access-controlled storage.

In a custom GitHub workflow, add --github-summary to append a masked failure report to the job summary and configure caching/uploads explicitly. The supplied 0.8+ GitHub Actions integration handles references and optional report sharing in trusted private repositories; it is not a published Marketplace action. Its hosted cache/artifact flow has passed a private pilot, not a customer-project performance benchmark.

Install your team’s VS Code extension through Install from VSIX, then run Worked: Show Status. Select the trusted CLI executable when prompted. Opening the editor does not collect data.

The local versioned JSON API and supplied Node.js adapter share the same project history. There is no HTTP server.

worked api capabilities
worked api status --workspace /absolute/project/path
worked api alerts --workspace /absolute/project/path

For a compatible MCP client, use an absolute executable and workspace:

{
  "mcpServers": {
    "worked": {
      "command": "/absolute/path/to/worked",
      "args": ["mcp", "--workspace", "/absolute/project/path"]
    }
  }
}

MCP runs over standard input/output and is read-only by default. It exposes status, collection scope, history, reports, alerts, fixed incidents, activity and the analysis prompt. Optional history writes require server opt-in and per-call confirmation; this is not permission to execute tests or apply fixes. Your MCP client may share results with its model provider.

Collectors recognize dpkg, RPM and pacman; this does not make one binary compatible with every Linux distribution. Unavailable sources produce visible collection warnings.

Worked stores structured metadata and fingerprints locally, not raw watched-file contents or environment-variable values. Approved Node manifests and version files supply selected requirements, pins and npm lockfile versions; other supported lockfiles supply fingerprints. These describe declarations, not proof of installed dependencies. Secret-like files are excluded by default. Inspect and choose collectors:

worked scope
worked scope disable processes
worked scope enable processes

No telemetry or automatic uploads. Reports are masked, not guaranteed anonymous: branch names, packages and other identifiers may remain. Review them before sharing. CI artifacts and MCP results follow your chosen provider’s access rules.

Alerts preserve the last manual comparison and the working baseline. The latest 100 are indexed by worked alerts; report files remain until explicitly removed.

Incidents retain up to 100 selected changes and 100 failing test names, with explicit omissions. Reviews and verification outcomes are append-only. Incidents and activity survive snapshot cleanup, are included in backups and have no automatic expiry. Old reports, passing runs and monitoring alerts are not converted into incidents.

worked backup --output /absolute/path/history.sqlite3
worked support --output /absolute/path/worked-support.json

Use new filenames. Support exports contain operational metadata, not your inventory or license contents; nothing is sent automatically.

Update from a verified package with bash scripts/install-cli.sh --replace. The previous executable and history are retained. Signed-release packages also support worked update --from /absolute/path/extracted-package.

bash scripts/uninstall-cli.sh removes the executable and owned schedule, not history, backups or shell configuration.

Send your SVÄRTA AI contact the command, warning and reviewed support summary—not your raw database or license file.