[baokhang83]

FluencyLoop retains useful reasoning, but agents do not consistently use it when designing the next change. This plan proposes a 1.0 loop that grounds recommendations in product intent, principles, and prior decisions while preserving human fluency and uninterrupted work.

The interaction contract is: use retained reasoning automatically, explain consequential choices at the right depth, recommend a path, continue within agreed intent, ask when human judgment is needed, and update the shared model at natural boundaries.

The plan explicitly protects FluencyLoop's standing principles:

• No separate proposal approval for already-authorized work. Batch genuine unresolved choices with a recommendation and reuse settled answers.
• No mandatory brief, historical cleanup, automatic unfamiliarity probe, or comprehension checkpoint. Requested guided learning remains supported without quizzes.
• Establish enough product direction before the first substantial plan/feature design: reuse known audience/outcome/exclusions and ask only consequential gaps with a recommendation. Later work reuses answers and asks only novel product forks; review recommends direction revisions when evidence challenges an assumption. A completed brief remains optional, and init/routine fixes do not trigger vision questionnaires.
• Capture rationale once, reuse unchanged context, record outcome deviations at natural closure, and distill once at wrap-up.
• Run T1 as an observed pilot with existing artifacts before committing to larger runtime/schema work. Use its evidence to narrow or defer scope.
• Keep the optional site focused on understanding and challenging the design, with passive uncertainty/provenance rather than compliance scores or repair obligations.
• Read existing customer knowledge without mandatory data migration. Startup and read paths must not alter knowledge files or bulk-append reconstructed history; old and new records coexist, and historical enrichment is explicitly requested and scoped. T2 owns reader compatibility from the start; T10 verifies it across runtimes, skills, and site.

Retains the FluencyLoop name and proposes “Stay fluent as people and agents evolve the product.” The 12 tasks in the 1.0 milestone, issues #177–#188, cover the pilot, justified runtime changes, workflows, site, compatibility, full regression evaluation, and release. The issue scopes and acceptance criteria reflect this contract; T10 now explicitly requires zero-write reads and no mandatory data migration. T1, T5, T6, and T11 explicitly cover initial framing without a prewritten brief, reuse of direction, and evidence-backed revision conversations. Version and positioning remain proposals. This PR changes planning documentation only.

Validation: checked the diff for whitespace errors and verified all 12 plan/issue titles, numbers, dependency links, milestone assignments, and published issue bodies. Runtime tests were not run because no runtime code changed.