SKILL.md
Verify CLI
Run a comprehensive set of smoke tests against the mpak CLI to catch regressions before publishing. This skill builds the CLI, then exercises every command category against the live production registry.
Prerequisites
Before running tests, ensure the working directory is the mpak monorepo root (apps/mpak). Build the CLI and SDK fresh:
pnpm --filter @nimblebrain/mpak-sdk build && pnpm --filter @nimblebrain/mpak buildAll test commands below should be run from packages/cli/ using node dist/index.js.
Test Procedure
Run each test group below. For every command:
- Execute the command
- Verify the exit code is 0 (unless testing error cases)
- Verify the output matches the expected pattern described
- Record PASS or FAIL with details
If any test fails, stop and report the failure with full output before continuing.
1. Version and Help
| # | Command | Expected |
|---|---|---|
| 1.1 | node dist/index.js --version |
Prints a valid semver version (not "unknown") |
| 1.2 | node dist/index.js --help |
Shows usage with commands: search, run, bundle, config, completion |
| 1.3 | node dist/index.js bundle --help |
Shows subcommands: search, show, pull, run |
2. Search
| # | Command | Expected |
|---|---|---|
| 2.1 | node dist/index.js search "mcp" |
Prints "Found N result(s)" and a "Bundles (N):" section with a single table |
| 2.2 | node dist/index.js search "mcp" --json |
Valid JSON with a results array and a totals object ({ "bundles": N }) |
| 2.3 | node dist/index.js search "mcp" --limit 2 |
Shows at most 2 results |
| 2.4 | node dist/index.js search "xyznonexistent999" |
Prints "No results found" message |
Verify that the Bundles table has columns: NAME, VERSION, TRUST, DESCRIPTION.
3. Bundle Commands
Use @nimblebraininc/echo as the test bundle (known to exist with trust level L0).
| # | Command | Expected |
|---|---|---|
| 3.1 | node dist/index.js bundle search "echo" |
Returns at least 1 result including @nimblebraininc/echo |
| 3.2 | node dist/index.js bundle show @nimblebraininc/echo |
Shows name, version, type, trust level (L0), provenance, statistics, versions, platforms |
| 3.3 | node dist/index.js bundle show @nimblebraininc/echo --json |
Valid JSON with name, latestversion, certificationlevel fields |
| 3.4 | node dist/index.js bundle pull @nimblebraininc/echo -o /tmp/mpak-test-echo.mcpb |
Downloads file to /tmp/mpak-test-echo.mcpb, file exists and is non-empty |
| 3.5 | node dist/index.js bundle show @nimblebraininc/doesnotexist |
Exits with error containing "not found" (case-insensitive) |
After test 3.4, clean up: rm -f /tmp/mpak-test-echo.mcpb
4. Config Commands
These test the local config manager at ~/.mpak/config.json.
| # | Command | Expected |
|---|---|---|
| 4.1 | node dist/index.js config set @test/smoke-test smokekey=smokevalue |
Succeeds with confirmation message |
| 4.2 | node dist/index.js config get @test/smoke-test |
Shows smoke_key with masked value |
| 4.3 | node dist/index.js config get @test/smoke-test --json |
Valid JSON containing smoke_key |
| 4.4 | node dist/index.js config list |
Lists @test/smoke-test in output |
| 4.5 | node dist/index.js config clear @test/smoke-test smoke_key |
Succeeds with confirmation |
| 4.6 | node dist/index.js config clear @test/smoke-test |
Succeeds, clears entire package config |
| 4.7 | node dist/index.js config get @test/smoke-test |
Shows no configuration or empty result |
5. Error Handling
| # | Command | Expected |
|---|---|---|
| 5.1 | node dist/index.js search |
Exits with non-zero code, shows error or help about missing argument |
| 5.2 | node dist/index.js bundle invalid |
Shows help or error about unknown command |
Reporting
After all tests complete, print a summary table:
## CLI Smoke Test Results
| Group | Passed | Failed | Total |
|--------------------|--------|--------|-------|
| Version & Help | N | N | 3 |
| Search | N | N | 4 |
| Bundle Commands | N | N | 5 |
| Config Commands | N | N | 7 |
| Error Handling | N | N | 2 |
|--------------------|--------|--------|-------|
| TOTAL | N | N | 21 |
Overall: PASS / FAILIf any test failed, list each failure with its test number, command, expected result, and actual output.
The CLI is ready to publish only if all 21 tests pass.