A LIVING FIELD GUIDE · UPDATED EVERY 10 HOURS
Better systems.
Shorter loops.
Software that ships.
A practical notebook for building a software factory: agent harnesses, feedback loops, quality gates, and the tools worth bringing into your workflow.
Explore the latest ↓Evidence before confidence.01 / THE RADAR
Useful signal.
Less noise.
Primary sources, practical takeaways, and a clear record of what actually changed.
Loading the field notes…
RESEARCH LOG
What changed
02 / THE PLAYBOOK
A factory needs
a feedback loop.
Move one coherent slice from intent to evidence. Keep the decision to release explicit.
01 Frame the work ↗
Define the problem, acceptance behavior, constraints, and authorized changes. Inspect the repository and establish a baseline before editing.
TASK.md → acceptance criteria02 Build one slice ↗
Choose the smallest end-to-end result. Use an isolated branch when needed and preserve unrelated work. Keep the implementation aligned with the acceptance criteria.
Inspect → plan → implement03 Verify & review ↗
Test observable behavior, inspect the diff, and record proof. If a check is unavailable, say so. Stop after three failed repair cycles and identify the blocker.
EVIDENCE.md → measured outcomes04 Approve, ship, learn ↗
Present the exact change, impact, and rollback before requesting release approval. Verify production after release. Keep only lessons supported by real failures.
Approval → release → retrospective03 / THE TOOLBOX
Capabilities with
a reason to exist.
Reviewed skills installed for this workflow. New discoveries are recommendations until compatibility and permissions are checked.
Codex and OpenClaw can share compatible SKILL.md workflows. Plugin packages and account connections are verified separately. Newly discovered executable extensions require review and approval.
BUILT TO KEEP LEARNING
The loop is the product.
Research refreshes every ten hours. Meaningful updates arrive on Telegram with a link back here.
Back to the top ↑