Tester Agent

Portable Cursor skill package. Routes a backlog-item test run through three subskills. Report artefacts match the CLI (src/reports.ts) — see reports.md.

When to use

• User asks to test a backlog item / story.
• User asks for the tester-agent workflow (plan → write tests → run).
• User wants a repeatable tester process without the CLI app.

Read first

1. workflow/template.md — copy the checklist and fill it in.
2. reports.md — required filenames and formats (same as CLI).
3. test-design.md — what Fase 1 must put in the test plan.
4. Project overrides in .tester-agent/ if present; else bundled templates/workflow/.
5. Prompt shapes in templates/prompts/.

Do not edit files under templates/ in this skill when adapting a project — copy overrides into .tester-agent/ instead.

Effort

This skill cannot switch the Cursor model or the effort setting. Stay on the model the user already selected. For the steps below, reason at high effort before writing anything: consider the backlog, the applicable code, and alternatives, then choose.

Say the effort in the progress line when a high-effort step starts, for example [fase 1] effort: high.

Zichtbare voortgang

The user is often in another chat or another part of the app. Say what this run is doing in the chat, and append the same line to $REPORT_DIR/progress.log. Do this at the moment it happens, not in a summary at the end.

msg="[fase 1] geschreven: .tester-agent/reports/${RUN_ID}/test-plan.md"
printf '%s\n' "$msg" | tee -a "$REPORT_DIR/progress.log"

Then repeat that line in the chat.

Required moments:

• start of a phase, including mode (pre-implementation or post-implementation)
• each file read for the plan or the tests (path only)
• each file written (path only)
• each test command, then its exit code

Keep each line to one path or one command. Do not paste file contents or test output into the progress log.

Parallel work

The user may keep editing other application code while this skill runs. That is expected.

• Leave every file you did not create for this test run untouched, including unrelated dirty application files.
• Do not treat those edits as a failure or policy_violation.
• Do not run git restore, git checkout, git reset, git clean, git stash, or any other command that reverts or discards working-tree changes. A failing test stays a failed or residual-risk result.

Workflow

Follow phases in order. After each phase, update the checklist in the workflow template.

Routing

1. Start → read tester-plan and execute Fase 1.
2. When test-plan.md exists and Fase 1 is done → read tester-write-tests and execute Fase 2.
3. When Fase 2 is done → read tester-run-tests and execute Fase 3.
4. Stop when report.md, report.json, and report.csv exist.

If the user asks for only a plan, run Fase 1 and stop.
If the user asks for only running tests, run Fase 3 (requires existing plan / commands).

Run id and reports

Use the CLI run-id format:

RUN_ID="$(node -e "const {randomUUID}=require('crypto'); process.stdout.write(new Date().toISOString().replace(/[:.]/g,'-')+'-'+randomUUID().slice(0,8))")"
REPORT_DIR=".tester-agent/reports/${RUN_ID}"
mkdir -p "$REPORT_DIR"

Final artefacts (same as CLI):

Skill handoff only (not a CLI file):

Always use the scripts under scripts/ so formatting stays identical to the CLI.

Subskills

• tester-plan
• tester-write-tests
• tester-run-tests

Install into another project

cp -a .cursor/skills/tester-agent \
      .cursor/skills/tester-plan \
      .cursor/skills/tester-write-tests \
      .cursor/skills/tester-run-tests \
      /path/to/project/.cursor/skills/

Optional: add project-specific .tester-agent/*.md overrides (leave skill templates/ untouched).