Skip to content

AI turns documents or topics into real, native PowerPoint decks—with native shapes, transitions and animations, data-backed charts and tables on demand, audio narration from speaker notes, and support for your own .pptx templates. · by Hugo He

fix(prompts): calibrate text width once on P01, then estimate by font

hugohe3f28caf7ecb73
diff --git a/skills/ppt-master/references/executor-base.md b/skills/ppt-master/references/executor-base.md
index ee26bc9d..d2f08823 100644
--- a/skills/ppt-master/references/executor-base.md
+++ b/skills/ppt-master/references/executor-base.md
@@ -184,7 +184,7 @@ Mechanical repetition comes from reusing one carrier and topology without a page
- **Element grouping (Mandatory)**: wrap each logical Slide-local body unit in a descriptive, page-unique top-level `<g id>` with root-coordinate `data-pptx-bounds="x y width height"`; a helper-authored preset atom stays top-level with `data-pptx-frame` and no bounds; nested groups need none. Give a root background image or full-canvas scrim/decoration rectangle a stable `id` plus `data-pptx-role="background"` / `"decoration"` instead of a wrapper. Thresholds, exemptions, and the Morph staging marker: [`shared-standards-core.md`](./shared-standards-core.md) §4.3.
- **Reference — nested edit groups**: a top-level group may contain descriptive nested `<g>` groups for meaningful subunits; they need no bounds and create no animation step, with no default depth or quota.
-- **Default — bounds are the module zone, not a glyph box (may skip when no text is estimable)**: make each zone as generous as the canvas and siblings allow without overlap. An untransformed line spans `y − 0.85 × font_size` to `y + 0.35 × font_size`; width uses the shared per-run estimate with headroom, callable as `python3 ${SKILL_DIR}/scripts/text_measure.py measure|wrap|box` — one calibration per role, later calls only for lines near a limit (batch via `--stdin`). When text does not fit, first expand a zone with unused space, then reflow or adapt; larger bounds never repair off-canvas text.
+- **Default — bounds are the module zone, not a glyph box (may skip when no text is estimable)**: make each zone as generous as the canvas and siblings allow without overlap. An untransformed line spans `y − 0.85 × font_size` to `y + 0.35 × font_size`. **Width is calibrated once, then estimated**: on P01 measure one representative line per role with `python3 ${SKILL_DIR}/scripts/text_measure.py measure` (batch via `--stdin`) and keep "≈ N chars per W px" per role; every later page sizes its zones from that per-font arithmetic (characters × the role's calibrated width, plus its headroom) — write the sentence first, then fit the zone to it; no per-page measurement, no line-by-line tool calls, and never trim wording to satisfy an estimate. Reach for `measure|wrap|box` again only for a role or size that was never calibrated. When text does not fit, first expand a zone with unused space, then reflow or adapt; larger bounds never repair off-canvas text.
- **Spec adherence**: binding color, canvas, typography, identity, resource, and template anchors hold; layout and other References apply under §2.1 without becoming locks.
- **Template structure**: inherit the native framework only for `template_reuse_scope: mirror|layout`; `style` uses the flat route.
- **Main-agent ownership**: SVG generation runs in the main agent, never a sub-agent — pages share upstream context for cross-page continuity.
diff --git a/skills/ppt-master/workflows/generate-pptx.md b/skills/ppt-master/workflows/generate-pptx.md
index ec7cddd4..60bfb9cc 100644
--- a/skills/ppt-master/workflows/generate-pptx.md
+++ b/skills/ppt-master/workflows/generate-pptx.md
@@ -263,7 +263,7 @@ gate-signal: method=<rule resolved, or none> | page-local=<count> | not-exercise
| Signal | Reading |
|---|---|
-| Two or more issues share a category and direction | Method-level bias — resolve to the authoritative rule before P02 (for text extents, the shared estimator `python3 ${SKILL_DIR}/scripts/text_measure.py measure|wrap|box`: calibrate each role once, measure only lines near a limit, batch via `--stdin`); a correction fitted to the observed offset patches only this sample |
+| Two or more issues share a category and direction | Method-level bias — resolve to the authoritative rule before P02 (for text extents: fix the P01 per-role calibration from `python3 ${SKILL_DIR}/scripts/text_measure.py measure`, then every later page estimates by that arithmetic and measures nothing); a correction fitted to the observed offset patches only this sample |
| One isolated issue tied to this page's structure | Page-local — fix and continue |
| A recurring element appears for the first time (furniture, caption format, section numbering, accent discipline) | It will be copied to every later page — confirm its semantics now |
diff --git a/skills/ppt-master/workflows/profiles/quick-generate.md b/skills/ppt-master/workflows/profiles/quick-generate.md
index 51853850..f13535f4 100644
--- a/skills/ppt-master/workflows/profiles/quick-generate.md
+++ b/skills/ppt-master/workflows/profiles/quick-generate.md
@@ -180,7 +180,7 @@ Read the execution core together, never file by file: [`shared-standards-core.md
**PPTX structure**: speed never flattens template structure. Free design and Brand/Style-only author flat Slide-local SVG with one root `data-pptx-page-role` (`cover` / `toc` / `section` / `content` / `ending`) and no Master/Layout/layer/placeholder metadata. When Layout or Deck owns structure, every page is a complete structured Slide SVG that preserves or deliberately adapts the prototype's root identity, fixed layers, and slots with current content on top — all-or-none across the roster, every reused Layout repeating an identical fixed-layer/slot contract, a new Layout allowed under the selected Master when the application paragraph calls for adaptation, ownership never inferred from repeated geometry, and `data-pptx-page-role` omitted. A Style never strips structure; only an explicit instruction to use the workspace as visual language permits flat output.
-**Typography**: name a concrete target-installed/approved family under `shared-standards-core.md` §4.1, never a lock or the host's fonts. Before P01 measure one representative line per recurring role with `python3 ${SKILL_DIR}/scripts/text_measure.py measure` and keep "≈ N chars per W px" in context (the checker adds wrapping headroom, so characters × font size underestimates by roughly 15%); measure only lines near a limit, batched through `--stdin`, and `wrap` only a genuinely long paragraph.
+**Typography**: name a concrete target-installed/approved family under `shared-standards-core.md` §4.1, never a lock or the host's fonts. Before P01 measure one representative line per recurring role with `python3 ${SKILL_DIR}/scripts/text_measure.py measure` (batched through `--stdin`) and keep "≈ N chars per W px" in context (the checker adds wrapping headroom, so characters × font size underestimates by roughly 15%); every later page sizes zones from that per-font arithmetic — write the sentence first, fit the zone to it, and never trim wording to satisfy an estimate; measure again only for a role or size never calibrated, and `wrap` only a genuinely long paragraph.
**Generation pacing**: hand-write the roster in order — P01 calibrates visual identity and cover expression, the first ordinary content page calibrates content geometry and carrier integration, neither becomes a reusable template — with no first-page checker or confirmation stop. A resolved motif follows its reuse mode (exact repetition for deliberate chrome, adaptive variation of scale, crop, density, position, or interaction otherwise). After every page exists, run the one final checker below; use other stages only when their capability is needed.