[1.0 T5] Recommend and continue within authorized intent in plan and feature workflows #181

Problem and outcome

Make retained reasoning improve the agent design process without adding a proposal ritual, approval menu, or compulsory learning checkpoint.

Interaction contract

Keep FluencyLoop. Proposed positioning: Stay fluent as people and agents evolve the product.

Use retained product reasoning automatically. Explain consequential choices at the right depth. Recommend a path. Continue within agreed intent. Ask when human judgment is needed. Update the shared model at natural boundaries.

The flagship goal/principles/precedents/tradeoffs reasoning is integrated into the existing design explanation at proportional depth. It is not a mandatory recital, proposal form, or new approval exchange. People must remain able to understand and challenge the design.

Scope

• Load or reuse relevant context automatically before consequential choices. Integrate the flagship reasoning into the existing design explanation rather than reciting every field before every edit.

• State the recommended approach and useful reason, then continue within authorized intent. Reuse settled choices and preferences, including an explicit user instruction that already revises a rule.

• Batch consequential unresolved questions with a recommendation. Wait for dependent work and continue independent work; never treat elapsed time as approval.

• Explicitly replace the current automatic blocking unfamiliarity probes and mandatory understanding confirmations. Unknown familiarity is not fluency; give useful concise context without forcing a checkpoint. Support deliberate teaching pauses when the person requests guided learning, preserve that preference, and never quiz.

• A missing brief or unrelated historical defect is advisory; use expressed intent when sufficient. Calibration changes depth, not the design choice or decision authority.

• Update both packaged skill surfaces consistently and keep rationale capture and wrap-up bounded.

• At the first substantial plan, or first substantial feature when no plan is needed, identify the audience, desired outcome, and important exclusions already established. Recommend a framing and ask one batched set of consequential unanswered questions before choosing a dependent design. No fixed three-question ritual.

• For subsequent initiatives and product forks, reuse the established direction and question only the new uncertainty that changes the product or experience. A new session or absent brief does not reset agreements.

• The agent distills agreed direction into the brief at a natural artifact boundary; no user-authored template, separate approval of an already-agreed summary, or automatic reconstruction on project opening. Keep unanswered fields unknown.

• Keep init as setup and routine fixes lightweight. A completed brief is optional; resolving product uncertainty that determines design is necessary. Reuse existing intent even when it has not been written into a brief.

Acceptance criteria

• An authorized consequential task receives a proportionate recommendation and continues without separate proposal confirmation.

• A real unresolved fork receives a recommended path and batched question before dependent implementation.

• An explicit choice is not re-asked because a record lacks human-approved status.

• Unknown familiarity does not trigger a compulsory probe/check; requested guided learning still pauses usefully without quizzes.

• Routine edits avoid full flagship recitals; missing documentation does not force a strategy/cleanup conversation.

• Observed fresh sessions show relevant prior reasoning used, not irrelevant citations added to satisfy a template.

• With partially specified first-substantial-work intent, the agent asks a recommended, focused product question before selecting the design it affects.

• With fully specified intent, the agent proceeds without a product questionnaire. Later work reuses the same audience/outcome/exclusions and asks only novel material forks.

• The agent writes a faithful concise brief from agreed answers without a second ratification exchange; unagreed direction remains a proposal.

• Init, routine fixes, new sessions with settled direction, and document absence alone do not trigger vision elicitation.

Validation

Check skill package parity and observe Claude Code/Codex sessions using T1 scenarios. Count forced pauses, repeated questions, and explanation size separately. Include a user-directed principle revision, unknown domain, and deliberately irrelevant goal.

Exercise first substantial plan and direct feature entry with both partially specified and sufficient intent, then a fresh session reusing the answers and a later initiative with one novel product fork. Check both packaged skills; verify questions occur before dependent design and do not become a document-completeness gate.

Starting points

plugins/fluencyloop/skills/{fluencyloop,plan,feature}/SKILL.md; claude-skills equivalents; tests/plugin.bats and PowerShell mirror

Constraints

• Target dev; main distributes releases. No runtime implementation occurs in the planning PR.

• T1 is an observed pilot using existing artifacts before new infrastructure. Use its evidence to retain, narrow, or defer later scope.

• Recommend and continue within authorized intent; ask with a recommendation only for consequential unresolved judgment. Batch independent questions, reuse settled answers, and wait for dependent work. Silence is not approval.

• Absent briefs, historical link gaps, stale prose, and unreviewed metadata alone are advisory. Do not require onboarding, cleanup, or retrospective approvals to work.

• Unknown familiarity is not fluency and does not automatically require a blocking probe/check. Support requested guided learning without quizzes; keep calibration/preferences private. T5 explicitly updates inherited pause rules.

• Capture rationale once, link mechanical views, record outcome deviations at natural closure, distill once at wrap-up, and reuse unchanged context.

• Preserve append-only history/provenance; Bash and PowerShell implement the same contract. Core remains Node-optional; no new database, hosted service, or vector search.

• No brief prerequisite does not waive resolving product uncertainty that determines design: establish enough direction early, reuse agreements, and reopen only for a material fork or relevant new evidence.

Dependencies and size

• Task: T5
• Relative size: M (not a calendar estimate)
• Depends on: T3: [1.0 T3] Assemble bounded design context in Bash and PowerShell #179, T4: [1.0 T4] Link existing design rationale and capture outcome deviations once #180
• Milestone: https://github.com/baokhang83/fluencyloop/milestone/3
• Plan: draft source
• Planning PR: docs: propose FluencyLoop 1.0 shared product direction and design #189