lcs-explore

LCS Explore

Shared Contract

Follow ../lcs-shared/contract.md for:

• folder conventions,
• OKF frontmatter,
• artifact writing safety,
• Source of Truth rules,
• stable IDs,
• Handoff format.

Chain of Truth level: Light.

Purpose

• Clarify user intent before PRD creation.
• Explore meaningful alternatives and trade-offs.
• Resolve material uncertainty through structured interactive brainstorming.
• Distinguish the user's initial request from the actual problem, the refined intent, and the eventual solution direction — do not treat a proposed solution as the problem definition.
• Discover and refine user intent through the interview rather than assuming it upfront.
• Explore solution alternatives proportional to complexity, ambiguity, and risk; prefer the simplest approach that adequately solves the validated problem within known constraints.
• Record decisions, assumptions, risks, and stable source requirements.
• Persist the result as: .lcs/work-items/{timestamp}-{slug-work-item}/explore.md .lcs/work-items/{timestamp}-{slug-work-item}/intent.md
• Do not create PRD, tasks, production code, or patches.

Required References

Before starting an interactive explore session, read:

references/interview-protocol.md

When generating explore.md, read:

../lcs-shared/templates/explore.template.md

When generating intent.md, read:

../lcs-shared/templates/intent.template.md

Do not load the artifact templates during normal interview rounds unless the artifact is about to be generated.

Workflow

1. Establish a concise work name from the user's intent.
2. If .lcs/state.md exists, read it before continuing.
3. Blank work reuse: If a work item is already selected (current_work is set) with phase: new, no primary artifact exists in its directory, and the user is clearly continuing that same named work, reuse the existing ID/path. Do not generate a second timestamped directory. Write explore.md into the existing path and set phase to explore using the synchronization rule.
4. If no work item is selected or the user is starting genuinely new work, create/register a new work item before or together with the state transition.
5. Assess the topic complexity.
6. Recommend an Explore Level:
   • Easy
   • Medium
   • Hard
   • Auto
7. Let the user choose the level.
8. If the user's manual level is materially lower or higher than recommended:
   • explain the mismatch briefly,
   • warn about the likely consequence,
   • ask whether they want to keep their choice or switch.
   • Never override a manually selected level without confirmation.
9. Run the interview using references/interview-protocol.md.
10. Ask 3 related high-value questions per round. Early rounds should prioritize collapsing uncertainty about the actual problem and intended outcome before locking onto a solution — if the user's initial request already names a solution (e.g. "add Excel export"), ask why it's needed and what problem it solves before treating it as a requirement.
11. After each round:

• interpret the answers,
• separate resolved decisions from unresolved questions,
• update the working understanding of problem, intent, and solution direction as distinct things,
• provide a 1-3 line recap,
• update remaining uncertainty,
• update approximate progress.

1. Continue until:
   • the work becomes PRD-ready,
   • the selected question budget is reached,
   • or a blocker prevents responsible progress.
2. Question count does not determine PRD readiness.
3. When relevant (non-trivial ambiguity, complexity, or risk), investigate existing system capabilities before proposing new solutions, and note relevant alternatives and trade-offs. Skip this for simple, low-risk changes (e.g. a label change) — do not force solution exploration where the solution is obvious.
4. Before finalizing, restate the understood Problem, Proposed Outcome, Constraints, and Out of Scope back to the user and give them a chance to correct material misunderstandings. Do not treat silence as approval for material decisions.
5. At the end of the main exploration, let the user choose whether to:
   • finish and hand off to PRD,
   • add one round,
   • deep-dive a specific area,
   • increase the Explore Level.
6. When the user finishes exploration:
   • read ../lcs-shared/templates/explore.template.md,
   • synthesize the complete explore.md artifact,
   • persist explore.md,
   • read ../lcs-shared/templates/intent.template.md,
   • synthesize intent.md from the confirmed Problem, Proposed Outcome, Affected Users/Systems, Constraints, Out of Scope, and Open Questions,
   • set Intent Status to confirmed if the readiness rules below are met, otherwise refining (or blocked if a material ambiguity prevents downstream work),
   • persist intent.md,
   • update .lcs/state.md,
   • hand off to lcs-toprd.

Exploration Priorities

Prefer questions that collapse the largest material uncertainty first.

Typical dimensions include:

• problem and intended outcome,
• users and actors,
• must-have behavior,
• existing-system constraints,
• architecture and data flow,
• integrations,
• UX expectations,
• security and privacy,
• deployment/runtime,
• scale/performance,
• alternatives and trade-offs,
• explicit non-goals.

Do not mechanically ask every category.

Do not ask questions whose answers are already known from:

• the user's current input,
• previous rounds,
• existing Source of Truth artifacts.

Adaptive Behavior

After every round, reassess what still matters.

If an answer makes a future question irrelevant:

• drop the irrelevant question,
• do not re-ask it,
• adapt the next round.

If the user answers only part of a round:

• preserve resolved answers,
• mark only the unanswered material items unresolved,
• do not repeat the entire round.

If the user gives a custom answer instead of selecting A-D:

• accept it,
• interpret it faithfully,
• do not force it into an existing option.

Decision Discipline

Keep agreed decisions separate from:

• assumptions,
• recommendations,
• unresolved questions.

Every agreed decision that implies a product, behavior, interface, data, security, or operational requirement must enter the Decision Ledger as an atomic stable SRC-### entry.

Rules:

• one question may create multiple SRC-### entries,
• one round may create zero, one, or many SRC-### entries,
• do not invent agreement,
• preserve existing IDs,
• mark unresolved assumptions [verified] or [unverified].

PRD Readiness

Treat exploration as PRD-ready only when:

• intended outcome is clear,
• major constraints are known,
• meaningful options and trade-offs have been considered,
• material risks and assumptions are visible,
• unresolved questions are non-blocking or explicitly carried forward,
• agreed requirements are captured as stable SRC-### entries.

A completed question budget does not automatically mean PRD-ready.

If the main budget ends while important uncertainty remains:

• state the remaining uncertainty,
• report current PRD readiness,
• recommend another round or focused deep dive.

Intent Readiness

Treat intent.md as confirmed only when:

• the Problem can be stated clearly and is distinct from any proposed solution,
• the Proposed Outcome is clear,
• Affected Users/Systems are sufficiently known,
• material Constraints are known,
• material Out of Scope items are identified,
• no remaining ambiguity materially blocks PRD work,
• the user had an opportunity to correct the agent's interpretation,
• no critical assumption is disguised as confirmed intent.

Not all Open Questions must be resolved — non-blocking questions may carry forward.

Use refining while the above is still incomplete, blocked when a material ambiguity prevents downstream work, and superseded only when a later Intent replaces this one.

Completion

When exploration finishes:

1. Generate explore.md using the canonical shared template.
2. Generate intent.md using the canonical shared template, reusing the confirmed Problem, Proposed Outcome, Constraints, and Out of Scope from the interview — do not restate the user's raw initial request as the Problem if Explore established a more accurate one.
3. Follow Artifact Writing Safety from ../lcs-shared/contract.md.
4. Update .lcs/state.md with:
   • current_phase: explore
   • current_work: {timestamp}-{slug-work-item}
   • work_items[state.current_work].phase = explore
   • work_items[state.current_work].updated_at = <current-ISO-timestamp>
   • last_session_note: <brief summary>
   • timestamp: <current-ISO-timestamp>
5. End with Handoff recommending lcs-toprd, including intent.md in the Source of Truth Bundle.

If filesystem or repository access is unavailable:

• provide the complete artifact in chat,
• mark file/state operations as not performed,
• never claim a file was read or written when it was not.

Chain of Truth

Level: Light

Expose:

• sources checked,
• assumptions,
• actions taken,
• verification status,
• risks.

Do not expose hidden chain-of-thought.

Chain of Truth Level

Level: Light