Ask the question the repository cannot answer
Separate discoverable facts from product decisions, ask consequential questions, and turn answers into a bounded brief.
Builds on: Bounded briefs and acceptance criteria
Practice workspaceY
12 PRACTICAL LESSONS
Better task judgment
Select useful context, recover safely, ask better questions and assess evidence.
Original Codex-focused lessons developed from the cclearn curriculum evaluation. Each follows Predict → Explain → Demonstrate → Practice → Verify → Reflect → Transfer. You can enter the unfamiliar transfer scenario first. Decisions are saved as recognition practice, separate from implementation mastery.
Choose the right existing pattern
Find a relevant precedent, verify that its constraints match, and reuse conventions without preserving a known flaw.
Builds on: Repository exploration and scope control
Verify a change when there is no test suite
Match a claim to a discriminating runtime or visual check, preserve its evidence, and report what remains unverified.
Builds on: Acceptance criteria and verification evidence
Review the reviewer’s findings
Treat findings as claims, investigate uncertainty and distinguish defects from preferences or unauthorized redesign.
Builds on: Diff review and evidence quality
Build a context packet that earns its place
Choose relevant observations, code and constraints while excluding secrets and unsupported diagnoses.
Builds on: Repository exploration and bounded briefs
Recover without erasing unrelated work
Inventory changes and side effects, preserve evidence, and choose a bounded recovery instead of assuming undo restores everything.
Builds on: Diff review and scope control
Prune instructions without losing the contract
Keep durable rules accurate and actionable; relocate task-specific detail without hiding necessary constraints.
Builds on: Task briefs, repository instructions and evidence review
Design a skill that is easy to discover
Define when a reusable workflow applies, separate method from task data, and test both matching and non-matching requests.
Builds on: Task briefs, repository instructions and evidence review
Make a verification failure actually block progress
Distinguish a request, an executed check and an enforced gate; require fresh, relevant evidence before dependent actions.
Builds on: Task briefs, repository instructions and evidence review
Follow the symbol before changing the code
Trace a behavior through aliases, callers and runtime registration; separate search results from evidence that a change is safe.
Builds on: Repository exploration, context selection and verification evidence
Measure the workflow, not the first draft
Compare time to an accepted result, account for rework and task differences, and keep claims proportional to the evidence.
Builds on: Bounded tasks, review and verification evidence
Check whether the advice fits your setup
Classify a technical claim by its source, date, environment and evidence; distinguish stale advice from an unverified local capability.
Builds on: Codex environments, repository inspection and permission boundaries