Direct, test, debug and supervise the agent that writes the code.
Finish with: A bounded agentic development environment with a task brief, operating manual, diagnostic harness and an independently verified agent repair of a deliberately broken application.
Progress stays in this browser. Export to keep a portable record.
BuildBotsBetterA step demonstration outline; use optional narration while you follow it.
This browser checks the evidence format and your checklist. It cannot inspect your computer or certify a live login.
You have recorded evidence for every step. Repeat the core task from a fresh input and export your results. Review the source, account and platform limits before using the workflow on real work.
Repeat the last product-specific checkpoint from a fresh practice input, preserve actual output, then complete the Build Bot Better guard and restore checks. Export the evidence.
Sources checked 2026-10-05. External tool interfaces and plan availability can change; check the linked official source and local --help.
Source-owned text for offline reading and retrieval. Command success still requires actual local evidence.
Download the resource ZIP at the top of this Build. Extract it into a new folder you own (for example C:\BBBPractice on Windows, or ~/bbb-practice on macOS/Linux). Open a terminal in the extracted folder. Run the version command for your platform. All subsequent commands assume this folder.
Expected: Python reports version 3.10 or newer and the terminal is in the extracted resource folder.
python --version
Get-Location
Get-ChildItempython3 --version
pwd
lspython3 --version
pwd
lsWrite the sequence books → Stack Overflow → search → ChatGPT → copied code/errors → assisted coding → coding agents → agentic workflow. For one recent task, list what you wrote/debugged manually and what you could direct/verify instead.
Expected: The map distinguishes implementer, assisted programmer and overseer.
Write human → objective → agent → code/tools → environment → evidence → agent → human oversight. Circle every place that can fail in the practice project.
Expected: The loop includes observable evidence and a human review boundary.
Adapt delegation.md with objective, scope, constraints, verification and evidence. Tell an installed signed-in agent to inspect app.py and verify.py, then propose one small change without editing yet.
Expected: The task brief has five concrete fields and the proposed files stay inside the practice folder.
Run verify.py yourself. Save the command, exit code and complete output. Compare this proof with a sentence claiming “all tests pass.”
Expected: VERIFY PASS is recorded with a command and exit code.
python verify.pypython3 verify.pypython3 verify.pyFill incident-template.md with working directory, Python version, input, expected result, actual output and one diagnostic. Ask the agent to identify missing observability before it suggests a fix.
Expected: The incident report can reproduce the symptom without private credentials.
Write AGENTS.md in the practice folder: objective, directory map, install/run/test commands, logs location, boundaries and stop condition. Keep verification in a dedicated file.
Expected: AGENTS.md contains reproducible run/test commands and explains the directory structure.
Write the sequence DEFINE → INSPECT → PLAN → EXECUTE → TEST → OBSERVE → DIAGNOSE → CORRECT → VERIFY → REVIEW. Use the agent for a docstring change, then run verification independently and inspect the diff.
Expected: Only the requested docstring changed; VERIFY PASS and a reviewed diff are retained.
python verify.pypython3 verify.pypython3 verify.pyRecord: stop retries, preserve stderr/logs, locate last checkpoint, inspect environment, reproduce minimally, form one hypothesis, delegate smallest repair, independently verify and review. Apply the plan to a deliberately missing practice input.
Expected: The recovery plan has eight steps and an observed failure/recovery pair.
Give the agent incident-template.md, app.py and verify.py. Request a diagnostic summary citing the exact file/line and failing assertion. Ask it to gather its own command output, then separate facts from hypotheses.
Expected: The diagnosis names a failing case, observed evidence and one testable hypothesis.
Make a three-row table: assisted (you choose each action), supervised (bounded task with review), autonomous (preapproved bounded loop with stop rules). Place code edits, test runs, external messages and production deploys into the appropriate policy.
Expected: The table distinguishes all three levels and requires explicit review for external actions.
Write the duties for this project: define outcome, constrain scope, provide tools/context, demand evidence, review diffs, protect data, monitor cost and decide whether to release. Assign an owner to every duty.
Expected: The job description names an accountable owner and check for each duty.
In this practice folder only, copy broken_app.py over app.py. Run verify.py and retain the repeated-input assertion failure. Do not change verify.py.
Expected: The repeated-input case fails before repair; the test remains unchanged.
Copy-Item -LiteralPath broken_app.py -Destination app.py
python verify.pycp broken_app.py app.py
python3 verify.pycp broken_app.py app.py
python3 verify.pyGive your signed-in agent delegation.md and the failure evidence. Ask it to diagnose and fix app.py only. Do not write the fix yourself. Review the minimal diff, independently run verify.py, and record the failure-to-pass pair.
Expected: VERIFY PASS follows a minimal agent-authored fix that preserves repeated inputs and leaves verification intact.
python verify.pypython3 verify.pypython3 verify.pyFill in onboarding-checklist.md for your agent. Record AGENTS.md, the tool’s user settings and memory folder, the assigned workspace, permissions, guard location, backup schedule, and a final proof task. Keep the checklist in your project.
Expected: The checklist contains a concrete path and owner for rules, memory, permissions, guard, backup, restore and final verification.
Run workspace_guard.py --self-test. Read the source: the target is resolved before deciding. The variable_target case points at the home folder but performs no filesystem operation. Compare this with a hook that sees only command text.
Expected: GUARD PASS reports inside allowed and profile, outside, variable-derived profile and workspace root denied; no deletion was executed.
python workspace_guard.py --self-testpython3 workspace_guard.py --self-testpython3 workspace_guard.py --self-testFor CLI tools, open the hook/permissions screen and confirm the configured guard is active. Use your tool’s documented pre-command hook protocol; the BBB working reference is linked in Sources. Run a harmless directory listing, then a guard dry-run targeting the profile. Do not execute a destructive profile probe. Grok Bot has no proven local shell guard here: use project-only tasks and reviewed actions, with backups.
Expected: A harmless command runs, the guard dry-run returns DENY, and the checklist records the actual hook protocol and active state or the Grok Bot limitation.
Connect a second backup drive. Windows: configure System Protection as additional recovery and use a scheduled copy/File History that explicitly includes hidden agent folders. macOS: configure Time Machine and verify dot folders are included. Linux: use a home backup such as restic/Deja Dup; Timeshift omits /home by default. Download the copy-only profile backup script for your platform, inspect it, and adapt the folder list to your tool. Record the last successful scheduled run; protect backup access because it may contain credentials.
Expected: A backup on another drive includes the tool’s user settings and memory folder and this project, and a successful schedule/time is recorded.
.\profile-backup.ps1 -Destination D:\Backups\llm-profilesbash profile-backup.sh /Volumes/YourBackup/llm-profilesbash profile-backup.sh /media/YourBackup/llm-profilesRun restore_drill.py with a separate backup directory. The script creates a new proof file, copies it, restores a second copy, and compares SHA256 hashes. Then restore one non-secret file from your actual agent memory backup into a separate scratch directory and open it; do not overwrite the live profile.
Expected: RESTORE PASS contains matching hashes; the actual memory file is readable after restoration into scratch.
python restore_drill.py --backup-root D:\Backups\bbb-practice --work-root practicepython3 restore_drill.py --backup-root /Volumes/YourBackup/bbb-practice --work-root practicepython3 restore_drill.py --backup-root /media/YourBackup/bbb-practice --work-root practice