Candidates and voting
Which topics should the book cover next? These are short proposals for new chapters. They are not accepted yet, and some may merge into existing patterns. The last two topics are proposed anti-patterns.
The full list, sources, overlaps, and current statuses are in CANDIDATES.md.
How to vote
Open “Discuss / vote” next to a topic and add a 👍 reaction to the opening post in the issue. You can support several topics. Comments describing a task where you need the pattern are especially useful.
Russian, English, and Spanish are welcome; participation requires a GitHub account. Votes indicate reader interest; inclusion remains an editorial decision.
| Topic | Identifier | Discussion |
|---|---|---|
| Visual specification — define the UI with a mockup and refine it using screenshots. | visual-specification |
Discuss / vote |
| Working example library — give the agent verified code to adapt and combine. | working-example-library |
Discuss / vote |
| Context forking — explore alternative solutions from the same prepared conversation state. | context-forking |
Discuss / vote |
| Behavior recovery — reconstruct an existing system’s specification with evidence for each conclusion. | evidence-linked-behavior-recovery |
Discuss / vote |
| Code walkthrough — follow an execution path with explanations and snippets from the repository. | linear-code-walkthrough |
Discuss / vote |
| Interactive explanation — understand an algorithm through inputs and step-by-step execution. | interactive-code-explanation |
Discuss / vote |
| Benchmark-guided optimization — compare the agent’s attempts and retain measured improvements. | benchmark-guided-optimization |
Discuss / vote |
| Approved scenarios — review readable inputs and expected results instead of many test functions. | approved-scenario-fixtures |
Discuss / vote |
| Independent acceptance — check the result with scenarios the implementing agent cannot reshape to fit its code. | held-out-acceptance-scenarios |
Discuss / vote |
| Mutation-guided test hardening — find faults that tests miss and strengthen their assertions. | mutation-guided-test-hardening |
Discuss / vote |
| Actionable diagnostics — return repair guidance alongside the violation message. | actionable-diagnostics |
Discuss / vote |
| Reviewable delivery — package the change, rationale, and verification evidence for review. | reviewable-agent-delivery |
Discuss / vote |
| Workflow learning — turn recurring corrections into shared instructions and checks. | feedback-flywheel |
Discuss / vote |
| Recovery points — preserve verified states so failed attempts can be undone in bounded steps. | verified-recovery-points |
Discuss / vote |
| Agent residue cleanup — remove temporary code and experimental leftovers before completing a task. | agent-residue-cleanup |
Discuss / vote |
| Instruction activation checks — determine whether the agent discovers a rule and applies it during the task. | instruction-activation-checks |
Discussion not opened yet |
| Guided manual procedure — have the agent prepare an interactive procedure for human actions and collect their results. | guided-manual-procedure |
Discussion not opened yet |
| Bounded retries — when progress stops, preserve findings and change the approach or hand off the task. | bounded-retry-escalation |
Discussion not opened yet |
| Approval fatigue — frequent requests turn considered approval into automatic clicking. | approval-fatigue |
Discuss / vote |
| Self-confirming tests — a test computes its expected answer using the very logic it should check. | self-confirming-tests |
Discuss / vote |