Structure a revisable analysis with ak:sequential-thinking

Use ak:sequential-thinking to analyze a complex problem in revisable stages. The Skill can expand or contract the working sequence, revisit an invalid assumption, compare branches, and verify hypotheses before presenting a final decision. It is a reasoning method, not an implementation or proof engine.

Choose ak:sequential-thinking for adaptive reasoning

Use ak:sequential-thinking when

• A problem has several dependent constraints or an unclear scope.
• New evidence may invalidate an earlier assumption.
• You need to compare two or three approaches and explain convergence.
• Debugging requires hypotheses, observations, and verification results to stay connected.
• A plan or architecture decision needs an auditable summary of assumptions, trade-offs, and unresolved questions.

Choose another workflow when

• The question is simple enough for a direct answer. Use ak:ask.
• You need broad option generation before analysis. Use ak:brainstorm.
• You need a repository implementation plan. Use ak:plan.
• You need to execute a change. Use ak:cook.
• You need repeated numeric optimization. Use ak:loop.

Prepare the problem and evidence

Before you start:

• Complete Onboarding, and confirm the Skill is available through Engineer Kit for the current runtime and scope.
• State the decision or diagnosis you need, not only the topic.
• Provide constraints, available evidence, rejected alternatives, and the authority boundary.
• Mark facts, assumptions, unknowns, and claims that require a test or source.
• Set a useful response bound such as maximum checkpoints, elapsed time, or branches. The Skill has no formal invocation flag for these limits.
• Keep secrets and sensitive personal data out of the reasoning record.

Run a bounded analysis

/ak:sequential-thinking "Choose a migration strategy for splitting the audit log from the primary database. Use at most 8 concise checkpoints, compare dual-write and change-data-capture branches, revise assumptions when evidence conflicts, and finish with a decision, verification plan, and unresolved questions."

/ak:sequential-thinking "Choose a migration strategy for splitting the audit log from the primary database. Use at most 8 concise checkpoints, compare dual-write and change-data-capture branches, revise assumptions when evidence conflicts, and finish with a decision, verification plan, and unresolved questions."

$ak:sequential-thinking "Choose a migration strategy for splitting the audit log from the primary database. Use at most 8 concise checkpoints, compare dual-write and change-data-capture branches, revise assumptions when evidence conflicts, and finish with a decision, verification plan, and unresolved questions."

Ask for visible checkpoints when they help review. For routine work, the Skill can apply the method implicitly and return only the conclusion, evidence, assumptions, and open questions. Neither mode exposes or guarantees a model’s private internal chain of thought.

Understand the reasoning markers

The initial total is a loose estimate. Expansion is appropriate when new constraints emerge; contraction is appropriate when a decisive finding makes later steps unnecessary. More checkpoints are not evidence of a better answer.

Observe the adaptive process

1. The Skill frames the question. It states the outcome, constraints, evidence, unknowns, and current scope estimate.
2. The Skill handles one concern at a time. Each checkpoint builds on prior context and says what evidence or decision comes next.
3. The Skill revises when necessary. A correction names the original claim, new evidence, and downstream impact.
4. The Skill branches selectively. Two or three plausible alternatives are compared before creating more branches.
5. The Skill verifies hypotheses. Tests, measurements, code evidence, or authoritative sources distinguish competing explanations.
6. The Skill checks its own progress. Circling, missing evidence, or a revision cascade triggers a strategy change rather than more repetitive prose.
7. The Skill converges or stops. It returns a decision with rationale and validation, or a blocked result naming the minimum missing evidence.

After a foundational revision, downstream checkpoints must be reassessed. Invalidated reasoning should not remain silently embedded in the final answer.

Use optional tracking carefully

The package includes optional Node.js helpers for deterministic validation, history, and display formatting. They are support files, not separate Skills or required runtime services.

The method itself needs no helper. Use support scripts only when durable local history or deterministic formatting is part of the requested deliverable.

Keep reasoning within its authority

Structured analysis does not authorize file edits, commands, network access, external research, deployment, or destructive actions. A verification step that needs any of those effects must follow normal project and user approval rules.

Do not turn missing evidence into confidence language. If a hypothesis cannot be tested within the available tools, state the gap and propose a bounded next check. Stop when further checkpoints repeat the same premise, exceed the user’s resource bound, or require new authority.

Verify the final result

A useful output should contain:

• The framed decision or diagnosis and its constraints.
• Facts, assumptions, and unresolved unknowns kept distinct.
• Revisions with the evidence and downstream impact that caused them.
• Compared branches with common evaluation criteria.
• Hypotheses and the evidence used to support or reject them.
• A final recommendation, confidence boundary, and verification plan.
• A blocked result when critical evidence is unavailable.
• Any optional local history artifact and its sensitivity boundary.

The run is complete when the requested decision is supported by available evidence and critical uncertainties are either resolved or surfaced. A FINAL label alone does not make the conclusion correct.

Troubleshoot or stop

Feed the verified decision to ak:plan, use ak:scenario to expand edge cases, or use ak:predict to stress-test a risky proposal.

Know the current limits

• The Skill structures analysis; it does not guarantee correctness, confidence, completeness, or an optimal decision.
• Visible markers are review artifacts, not access to private model reasoning.
• Hypothesis verification is limited by available evidence and authorized tools.
• The optional processor persists supplied text and does not provide encryption, access control, or cross-runtime synchronization.
• Runtime output shape and support-file location can differ after projection.
• Stable and beta contain identical ak:sequential-thinking source and helpers.