pptx-template-parsing

Name: pptx-template-parsing
Author: CxyZyr

15stars1forks

How to install pptx-template-parsing?

cxyzyr/pptx-template-skills/pptx-template-parsing

$npx -y skills add CxyZyr/PPTX-Template-Skills --skill pptx-template-parsing

Installs into the current project.

Summary

Use before adapting or debugging any PPTX template-driven deck. Parses a template into a portable spec.json contract: classified shape tree, slide roles, repeated card structures, per-slide fill_plan, text style/paragraph metadata, font availability, and optional renders. This is

Files · 1

View on GitHub

SKILL.md

1	# PPTX Template Parsing
2
3	Turn a `.pptx` template into a semantic `spec.json` that downstream generation can fill
4	mechanically — without the agent ever hand-reading shape indices off the slide.
5
6	This is the first half of the pipeline:
7
8	```
9	template.pptx ─[THIS SKILL]→ spec.json (+ summary.txt + renders/*.png)
10	│
11	▼
12	ppt-template-adaptation (authors content_plan.json, fills the deck)
13	```
14
15	## When to use
16
17	- A user provides a PPT template that will be reused (not redesigned).
18	- Before any generation: generation needs the `spec.json` slot map this skill produces.
19	- When you need to know, per slide: what role it plays, where the title/body/cards/icons/logo/images
20	live, and which slides repeat a card structure.
21	- When a generated deck has layout, style, overlap, or missing-slot problems. Check whether parsing
22	captured the right slots and text styles before adding generation fallbacks.
23
24	## Operating rule
25
26	The parser output is the contract. Do not hand-pick PowerPoint shape indices from memory, screenshots,
27	or a previous failed deck. If `fill_plan` is wrong, fix parsing/classification first, rerun this skill,
28	then regenerate with `ppt-template-adaptation`.
29
30	## What it produces
31
32	\| Output \| Purpose \|
33	\|---\|---\|
34	\| `spec.json` \| The machine-readable contract consumed by generation. See `references/spec-schema.md`. \|
35	\| `summary.txt` \| One line per slide: role + confidence + detected slots. Read this first. \|
36	\| `renders/*.png` \| One PNG per slide when rendering is enabled. Use for human/visual review when allowed. \|
37
38	## How to run
39
40	```bash
41	python3 skills/pptx-template-parsing/scripts/parse_template.py \
42	--template "PPT模板/大气人工智能科技感PPT模板.pptx" \
43	--out workspace/specs/ai_tech
44	```
45
46	Flags:
47	- `--no-render` — skip LibreOffice rendering (faster; do this for a quick structural pass).
48	- `--limit N` — only parse the first N slides (useful while iterating).
49	- `--dpi 110` — render resolution (default 110).
50
51	Rendering needs `libreoffice` + `pdftoppm` on PATH. If rendering fails the parser still emits
52	`spec.json` (renders are best-effort).
53
54	## Required workflow
55
56	1. Run `parse_template.py` against the original template, not a previously generated deck.
57	2. Read `summary.txt` and flag any low-confidence slide (`role_confidence < 0.5`) or unexpected slot
58	count, especially `contents` and repeated-card slides.
59	3. Read the relevant `fill_plan` entries in `spec.json`: `title`, `labels`, `body`, `cards`,
60	`images`, `charts`, `tables`, `logo`, and `visual_obstacles`.
61	4. For text-sensitive templates, inspect each slot's `style.paragraphs` / `title_style.paragraphs`.
62	Mixed-format boxes must preserve paragraph/run style instead of being flattened during generation.
63	5. Check `template.theme.font_availability`. Missing fonts explain render differences, but the PPTX
64	should still preserve the original font names in `style`.
65	6. Only after the slot map looks coherent, hand `spec.json` to `ppt-template-adaptation` and create a
66	`content_plan.json`.
67
68	## Debugging checklist
69
70	- Contents page: card count must match visible directory entries. If four entries parse as one large
71	card, fix card-list/group extraction here.
72	- Shared text boxes: if a single text box contains number/title/body with different styles, the parsed
73	style must include `paragraphs` with run-level sizes/colors.
74	- Alignment: preserve parser values such as `CENTER`, `LEFT`, and `DISTRIBUTE`. Do not reinterpret
75	WPS distributed alignment as ordinary center.
76	- Icon/image boundary: small semantic glyphs belong in `cards[].icon`; photos/screenshots belong in
77	`images[]`.
78	- Chart/table pages: if native chart/table slots are present, expose them as `charts[]` / `tables[]`;
79	do not treat them as ordinary text pages.
80
81	## How the classifier works (no single signal is trusted)
82
83	Each shape gets a `role`, a `role_confidence`, and the list of `signals` that fired, so its decision is
84	reviewable against the render. Roles are assigned in priority order — hard shape types (table/chart) →
85	real placeholders → stock placeholder text (`点击输入标题`) → font-size/geometry heuristics. Groups become
86	`card` only when they carry substantive text. Full rules: `references/classification-signals.md`.
87
88	Slide roles (cover / contents / section_divider / content / ending) and the per-slide `fill_plan` are
89	derived on top of the shape roles. Full rules: `references/slide-roles.md`.
90
91	## Reference read order
92
93	Read only what the task needs:
94
95	- `references/spec-schema.md` — required when consuming o

People who liked this also viewed

6 picks

Type

pptx-template-parsing

15stars1forks

View on GitHub

How to install pptx-template-parsing?

cxyzyr/pptx-template-skills/pptx-template-parsing

$npx -y skills add CxyZyr/PPTX-Template-Skills --skill pptx-template-parsing

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# PPTX Template Parsing
2
3	Turn a `.pptx` template into a semantic `spec.json` that downstream generation can fill
4	mechanically — without the agent ever hand-reading shape indices off the slide.
5
6	This is the first half of the pipeline:
7
8	```
9	template.pptx ─[THIS SKILL]→ spec.json (+ summary.txt + renders/*.png)
10	│
11	▼
12	ppt-template-adaptation (authors content_plan.json, fills the deck)
13	```
14
15	## When to use
16
17	- A user provides a PPT template that will be reused (not redesigned).
18	- Before any generation: generation needs the `spec.json` slot map this skill produces.
19	- When you need to know, per slide: what role it plays, where the title/body/cards/icons/logo/images
20	live, and which slides repeat a card structure.
21	- When a generated deck has layout, style, overlap, or missing-slot problems. Check whether parsing
22	captured the right slots and text styles before adding generation fallbacks.
23
24	## Operating rule
25
26	The parser output is the contract. Do not hand-pick PowerPoint shape indices from memory, screenshots,
27	or a previous failed deck. If `fill_plan` is wrong, fix parsing/classification first, rerun this skill,
28	then regenerate with `ppt-template-adaptation`.
29
30	## What it produces
31
32	\| Output \| Purpose \|
33	\|---\|---\|
34	\| `spec.json` \| The machine-readable contract consumed by generation. See `references/spec-schema.md`. \|
35	\| `summary.txt` \| One line per slide: role + confidence + detected slots. Read this first. \|
36	\| `renders/*.png` \| One PNG per slide when rendering is enabled. Use for human/visual review when allowed. \|
37
38	## How to run
39
40	```bash
41	python3 skills/pptx-template-parsing/scripts/parse_template.py \
42	--template "PPT模板/大气人工智能科技感PPT模板.pptx" \
43	--out workspace/specs/ai_tech
44	```
45
46	Flags:
47	- `--no-render` — skip LibreOffice rendering (faster; do this for a quick structural pass).
48	- `--limit N` — only parse the first N slides (useful while iterating).
49	- `--dpi 110` — render resolution (default 110).
50
51	Rendering needs `libreoffice` + `pdftoppm` on PATH. If rendering fails the parser still emits
52	`spec.json` (renders are best-effort).
53
54	## Required workflow
55
56	1. Run `parse_template.py` against the original template, not a previously generated deck.
57	2. Read `summary.txt` and flag any low-confidence slide (`role_confidence < 0.5`) or unexpected slot
58	count, especially `contents` and repeated-card slides.
59	3. Read the relevant `fill_plan` entries in `spec.json`: `title`, `labels`, `body`, `cards`,
60	`images`, `charts`, `tables`, `logo`, and `visual_obstacles`.
61	4. For text-sensitive templates, inspect each slot's `style.paragraphs` / `title_style.paragraphs`.
62	Mixed-format boxes must preserve paragraph/run style instead of being flattened during generation.
63	5. Check `template.theme.font_availability`. Missing fonts explain render differences, but the PPTX
64	should still preserve the original font names in `style`.
65	6. Only after the slot map looks coherent, hand `spec.json` to `ppt-template-adaptation` and create a
66	`content_plan.json`.
67
68	## Debugging checklist
69
70	- Contents page: card count must match visible directory entries. If four entries parse as one large
71	card, fix card-list/group extraction here.
72	- Shared text boxes: if a single text box contains number/title/body with different styles, the parsed
73	style must include `paragraphs` with run-level sizes/colors.
74	- Alignment: preserve parser values such as `CENTER`, `LEFT`, and `DISTRIBUTE`. Do not reinterpret
75	WPS distributed alignment as ordinary center.
76	- Icon/image boundary: small semantic glyphs belong in `cards[].icon`; photos/screenshots belong in
77	`images[]`.
78	- Chart/table pages: if native chart/table slots are present, expose them as `charts[]` / `tables[]`;
79	do not treat them as ordinary text pages.
80
81	## How the classifier works (no single signal is trusted)
82
83	Each shape gets a `role`, a `role_confidence`, and the list of `signals` that fired, so its decision is
84	reviewable against the render. Roles are assigned in priority order — hard shape types (table/chart) →
85	real placeholders → stock placeholder text (`点击输入标题`) → font-size/geometry heuristics. Groups become
86	`card` only when they carry substantive text. Full rules: `references/classification-signals.md`.
87
88	Slide roles (cover / contents / section_divider / content / ending) and the per-slide `fill_plan` are
89	derived on top of the shape roles. Full rules: `references/slide-roles.md`.
90
91	## Reference read order
92
93	Read only what the task needs:
94
95	- `references/spec-schema.md` — required when consuming o

People who liked this also viewed

6 picks

Type