agent-ide-cli-fixture-test

Agent IDE CLI Fixture Test

Use this skill to manually exercise the built agent-ide CLI on fixture projects and turn any failure into a tested fix.

Workflow

1. Inspect current state:

   • Run git status --short.
   • Confirm tests/fixtures has no leftover edits before starting.
   • If prior work changed source/tests, keep those changes; only clean fixture mutations created by the current case.

2. Build the CLI before manual tests:

   • Run pnpm build.
   • Use node bin/agent-ide.js ... for every manual CLI case.

3. For each case, run one CLI action against one fixture, then inspect exactly what changed:

   • Prefer mutation commands: rename, change-signature, move, deadcode.
   • Include read-only sanity cases: find-references, impact, cycles, search, and call-hierarchy.
   • Use fixtures under tests/fixtures/*; do not create unrelated throwaway projects.
   • Run git diff -- tests/fixtures/<fixture> immediately after each mutation.
   • Verify the final side effect, not only command success.

4. Validate the mutated fixture:

   • TypeScript fixture: run npx tsc --noEmit -p tests/fixtures/<fixture>/tsconfig.json when it has a tsconfig.json.
   • JavaScript fixture: run node --check on changed files and execute a fixture entrypoint when available.
   • If the command is read-only, verify git status --short -- tests/fixtures stays clean.

5. Restore before the next case:

   • Run git restore tests/fixtures/<fixture>.
   • Remove new files/directories from the case with scoped git clean -fd tests/fixtures/<fixture>/<path>.
   • Confirm git status --short -- tests/fixtures is clean.
   • Treat "reset fixture" as scoped restore/clean; do not use git reset --hard.

6. If a bug appears:

   • Keep the failing behavior concrete: command, fixture, diff, and compile/runtime error.
   • Add or update an E2E/integration test that fails for the bug.
   • Run the targeted test and confirm it is red.
   • Fix production code, not the fixture output.
   • Run the targeted test green, rebuild, then repeat the same manual CLI case.
   • Restore the fixture after retest.

7. Finish with validation:

   • Run pnpm test.
   • Run pnpm test:cli.
   • Run pnpm test:full when behavior changed or multiple CLI paths were touched.
   • Run pnpm typecheck, pnpm lint, pnpm build, pnpm validate:plugin, and git diff --check.
   • Report any command that was not run.

Case Coverage

Cover a mix of these cases when the user asks for "多測幾種 case":

• TS rename --at on a local symbol; expect only the selected scope to change.
• TS change-signature --rename, --add, --remove, and --reorder; expect body references and call sites to remain valid.
• TS move for a file, a member (src/file.ts:line), and a glob; expect imports and compilation to stay valid.
• JS rename --at; expect 1-based locations and correct import/call-site updates.
• JS move; expect ESM imports to keep required runtime extensions such as .js.
• JS/TS deadcode; expect deleted code to match unused symbols and changed files to parse/compile.
• Read-only commands; expect useful output and zero fixture diff.

Reporting

Report:

• Which fixture CLI cases were manually run.
• Which diffs were expected and verified.
• Which bugs were found and fixed.
• Which automated validation commands passed.
• Whether tests/fixtures is clean at the end.