liftosaur-program-creator（Liftosaur 程序创建者）

Liftoscript Program Authoring

Output contract

• Default to slash-style format for new lines: Exercise Name / set expressions / section: value / section: value.
• Keep one separator style per line; do not mix slash and freeform section placement.
• Preserve existing local style when editing unless the user asks to normalize.
• Use explicit section names for non-set data (progress:, update:, warmup:, superset:, used:, id:).

Operating rules

• Keep terminology consistent: use "exercise line", "set", "progress", "update", "state variable", "set variation".
• Prefer built-in progression (lp, dp, sum) before custom scripts.
• For scripting tasks, identify execution context first: progress: custom() vs update: custom().
• Use timer literals by context: 45s/120s in exercise-line sets, numeric timers in scripts (timers[5] = 20).
• Reject timed rep tokens in set syntax (3x60s); use numeric reps (3x60) when needed.
• Resolve units deterministically: nearby program text -> explicit user preference -> default kg.
• Do not silently mix kg and lb in one generated snippet.
• In exercise-line set syntax, % load tokens are percentages of 1RM (not training max).
• Resolve exercise naming from references/exercise-name-resolution.md against references/exercise-list.md.
• In full mode, place template lines (used: none) inside a day block; for week-level templates, anchor after ## Day 1, not directly after # Week N.
• Apply a validation loop: draft -> verify writable variables/scope/indexing -> correct -> finalize.
• Preserve user intent and existing structure; change only what is required.

Routing by task

• Basic syntax and set notation -> references/language-core.md
• Weeks, days, comments, labels, warmups, supersets, tags -> references/program-structure.md
• Built-in progression choice and arguments -> references/progression-builtins.md
• Reuse, repeat, templates, overrides -> references/reuse-and-templates.md
• progress: custom() authoring -> references/custom-progress-scripts.md
• update: custom() authoring -> references/update-scripts.md
• Read/write variables and indexing rules -> references/runtime-reference.md
• Liftoscript built-in functions -> references/builtins-reference.md
• End-to-end examples and fix patterns -> references/recipes.md
• Exercise name resolution -> references/exercise-name-resolution.md
• Known pitfalls and quick fixes -> references/common-failure-modes.md

Minimal workflow

1. Classify request: syntax, progression, reuse, custom script, or debug.
2. Identify mode before drafting: edit existing text vs generate new text; for scripts, choose progress vs update context.
3. Resolve units before drafting: nearby text -> user preference -> kg.
4. Read only the minimum references required (target 1-2 files from routing list).
5. Produce the smallest correct change.
6. Validate constraints (especially writable variables and index scope).
7. Return final program/script text and a short rationale.

Preflight validator (must pass before final output)

• Section separators are correct and consistent for each exercise line.
• Set and set-variation syntax is structurally valid (including grouped load/timer behavior when used).
• Script assignments target writable variables for the selected context.
• Script assignment literals use valid types for the target variable.
• Timer literals follow context rules (45s in exercise text, numeric in scripts).
• Rep tokens are numeric counts (no s suffix on reps; use 3x60, not 3x60s).
• Percent load tokens in exercise-line set syntax are interpreted as % of 1RM.
• Units are resolved from context/user preference or defaulted to kg.
• No forbidden unit mismatches in generated output.
• Exercise naming resolves against references/exercise-list.md (bare Exercise Name or explicit Exercise Name, Equipment).
• When user intent specifies equipment/variant, explicit Exercise Name, Equipment is an exact match in references/exercise-list.md.
• When user intent does not specify equipment, alias-only Exercise Name is an exact bare-name match in references/exercise-list.md.
• If exercise is not in built-in list, output remains valid and mentions manual add path in Liftosaur.
• In full mode output, template lines (used: none) are placed after ## Day 1 (or another explicit day), never directly under # Week N.
• All exercise line fields (used:, progress:, update:, warmup:, set expressions, timers) belong on one line separated by /; use \ only when a script body {~... ~} spans multiple lines.