Lab 4 - RobotCode

Teach your agent to work like a Robot Framework engineer who knows your setup: ask the project instead of grepping it, look up the installed libraries instead of guessing, and debug a failing test at a live breakpoint instead of re-running it blindly. All of it through the RobotCode command line. No server is involved. Keep the RobotCode cheat sheet open: which command answers which question, and the traps.

Module	4 - RobotCode for Agents
Time	25 minutes
Shop preset	clean
You need	Lab 0 done
You start from	main, plus your work from Labs 2 and 3 if you have it

Steps

1. Install the RobotCode plugin for your agent, from its marketplace. The version is recorded in SETUP.md:

   Claude Code	Codex	GitHub Copilot
   claude plugin marketplace add --scope project robotcodedev/robotframework-agent-plugins	codex plugin marketplace add robotcodedev/robotframework-agent-plugins	copilot plugin marketplace add robotcodedev/robotframework-agent-plugins
   claude plugin install --scope project robotcode@robotframework-agent-plugins	codex plugin add robotcode@robotframework-agent-plugins	copilot plugin install robotcode@robotframework-agent-plugins
   recorded in .claude/settings.json, so your fork remembers it	installed for your user	installed for your user

   Start a new session afterwards, so that the agent loads it.

2. Tie it to your context. Add one line to AGENTS.md, so that every session knows the plugin's commands apply here:

   Use RobotCode through uv (`uv run robotcode ...`) to discover tests and keywords, read library docs, debug
   failing tests and read results.

3. Ask a discovery question:

   Which tests of this repository carry the tag ui, and in which files are they? Ask the project; don't grep.

   Watch the tool calls: the answer should come from uv run robotcode discover, which resolves tests and tags the way Robot Framework does at run time. Reading the files gets half the story: tags can also come from settings and from the command line.

4. Ask a library question:

   Which keyword of the installed Browser library reads the state of a checkbox, and what are its arguments?

   The answer should come from uv run robotcode libdoc Browser: the version this repository pins, not the one the model remembers.

5. Debug a broken test. One of the two tests that fail on purpose is yours now:

   The test "WEB-002_AC-12 Handpicked Highlights" fails. Find out why with the RobotCode debugger: stop at the assertion that fails and look at the variables it compares. Don't guess from the failure message. Then tell me the cause before you change anything.

   The agent should run uv run robotcode robot-debug with a breakpoint, and report what it saw. Check its explanation against what the criterion says in openspec/specs/shop/catalogue. Then:

   Fix the test so that it verifies what the criterion says, remove its broken tag, and run it.

   The test passes. It is no longer broken on purpose, so it no longer carries the tag.

6. Explore with the REPL. The REPL is for flows no test covers yet. The price range filter (WEB-002_AC-8) is one:

   No test covers the price range filter yet. Explore it with the RobotCode REPL and a visible browser (HEADLESS set to False): open the products page through resources/shop.resource, set the range to $100-$300, apply the filters, and tell me which keywords and locators would work for a test. Don't write a test file.

   A browser window opens and follows the agent's steps. The REPL does not create its output directory, so opening the browser fails unless it exists: results/, which your earlier runs created, or one the agent passes with -d. The REPL is the right tool only here. An agent that opens the REPL to investigate the failing test of step 5 is misusing it: look out for that in the debrief.

Stretch

Ask a results question about the last run:

Of the tests tagged WEB-002 in the last run, which failed and why? Use the results, not output.xml.

It should use uv run robotcode results show with a tag filter.

Compare with the reference

When you are done, compare your result with the reference: what the rehearsal produced, why it is a good result, and the answers the debrief covers. Open it after the lab: it gives the answers away.

If your agent fails

Follow the recorded walkthrough of this lab. You can also run steps 3 and 4 yourself: uv run robotcode discover tests and uv run robotcode libdoc Browser list.