Tutorials and use cases

Verify your first coding session

Use when: you have connected an agent and want to confirm that Headroom is processing its requests.

1. Complete the quickstart and open a new terminal session.
2. Ask the agent: Read the project instructions, identify the test command, and run the existing tests. Report any failures.
3. Open Headroom's Activity view and look for requests or optimization records from this session.
4. Review the dashboard for the same period. Recorded requests confirm traffic; savings depend on the content processed.

Expected result: the agent completes the task and Headroom records activity. If there are no requests, check routing.

Reduce repetitive build and test output

Use when: the agent repeatedly reads long build logs or test results.

1. Enable RTK and start a new agent session.
2. Ask the agent to run the project's usual build or test command and identify failures.
3. Check that the summary contains the failures and diagnostics needed to continue. Inspect full output if details are missing.

Expected result: supported commands produce more compact output. Check Activity for available RTK records; installing the add-on alone does not establish savings.

Extract requirements from a document

Use when: a specification or report is supplied as a PDF or Office file.

1. Enable MarkItDown and start a new agent session.
2. Give the agent the local file path and ask: Convert this document to Markdown with MarkItDown. List the requirements and identify any tables or images that need manual review.
3. Compare the extracted requirements with the original document before using them to implement a change.

Expected result: the agent works from extracted text. Image-only pages or complex layouts may require additional review or extraction tools.

Investigate a repository

Use when: the agent reads many files to locate a function or trace a request.

1. Enable Serena for symbol lookup, or Codebase Memory for indexed relationships.
2. Start a new session and ask: Find the handler for this route, identify its callers, and read the relevant functions before proposing a change. Include the route or symbol name.
3. Check the cited files and symbols against the current repository. Review the agent's tool calls to confirm it used the add-on.

Expected result: focused code reads or indexed lookups replace some broad exploration. Added tools and indexing also have overhead; evaluate them on your own repository.

Implement against a library version

Use when: an integration depends on a specific library API.

1. Enable Context7 and start a new session.
2. Ask: Check the library version in this project. Use Context7 to look up the relevant API before implementing this change. Describe the change.
3. Review the documentation used and run the project's relevant tests.

Expected result: the implementation uses documentation appropriate to the dependency. If that version is unavailable, ask the agent to identify the gap before proceeding.

Review and improve recurring work

After several sessions, review project learnings for repeated command corrections or environment details. If output is too verbose, evaluate Ponytail or Caveman on a representative task.

Change one setting at a time. Check task quality as well as usage, and distinguish measured compression from estimated output reduction.