results-to-slides

Name: results-to-slides
Author: saidwivedi

23installs0forks

How to install results-to-slides?

saidwivedi/research-skills/results-to-slides

$npx -y skills add saidwivedi/research-skills --skill results-to-slides

Installs into the current project.

Summary

Automate the grunt work of making research presentations — discovers experiments from git/output folders, collects images and metrics, and organizes them into slides. Creates a slide-by-slide script for user approval, then generates slide markdown and editable PPTX.

Files · 1

View on GitHub

SKILL.md

1	# Results to Slides
2
3	Discover experiments, collect images/metrics, organize into slides. You provide the story —
4	this skill handles the grunt work. Output: slide markdown + editable PPTX.
5
6	## Core Rules
7
8	1. State what was done and what the result was. Do not editorialize. Never use:
9	"breakthrough", "key insight", "importantly". Just experiment + number.
10	2. Every slide MUST have a visual. No text-only slides (except pure tables). If a concept
11	has no experiment images, create a diagram. If no diagram makes sense, show dataset examples.
12	3. Images first, text second. Plan images for each slide BEFORE writing bullets. If you
13	can't find an image for a slide, restructure or merge it.
14
15	---
16
17	## Parallel Execution
18
19	Maximize use of the Agent tool. Whenever you have 2+ independent tasks, launch parallel agents.
20
21	- Phase 1: Launch agents in parallel to read CLAUDE.md, README.md, memory files
22	- Phase 2 (biggest win): Launch separate agents for git log, output folders, scripts, media
23	- Phase 3: Launch agents in parallel to read different experiment scripts and output folders
24	- Phase 5+6: Sequential (markdown must be written before converting)
25
26	---
27
28	## Important Paths
29
30	- Skill directory: `${CLAUDE_SKILL_DIR}`
31	- Converter: `${CLAUDE_SKILL_DIR}/md_to_pptx.py`
32	- Backgrounds: `${CLAUDE_SKILL_DIR}/backgrounds/`
33	- Theme CSS: `${CLAUDE_SKILL_DIR}/theme.css`
34	- Slide element reference: [slide_reference.md](slide_reference.md)
35
36	---
37
38	## Phase 0: Parse Arguments & Setup
39
40	The user provides: `/results-to-slides START_DATE END_DATE`
41
42	Arguments come as `$ARGUMENTS` containing two MMDD date strings (e.g., `0301 0308`).
43
44	Parse the dates:
45	- `$0` = start date (MMDD), `$1` = end date (MMDD)
46	- Infer year from system date. End date is INCLUSIVE (use end_date + 1 for `find` bounds)
47
48	If no arguments provided, use `AskUserQuestion`:
49	- header: "Date range"
50	- question: "What date range? Use MMDD MMDD format (e.g., 0301 0308)."
51	- options: ["Last week", "Last 2 weeks", "Last month"]
52
53	Output directory: `presentation/YYYY_MM_DD/` using the end date.
54
55	### Ask Presentation Preferences
56
57	Ask all preferences upfront before discovery, in a SINGLE AskUserQuestion call:
58
59	1. Background theme: "Light (Recommended)", "Warm", "Dark"
60	2. Emphasis areas: "What should the presentation emphasize? (e.g., dataset, model architecture,
61	results comparison, failed approaches)" — free text, helps prioritize slide count allocation.
62
63	\| Choice \| Background file \| `--theme` flag \|
64	\|--------\|----------------\|----------------\|
65	\| Light \| `slide_bg_light.png` \| `light` \|
66	\| Warm \| `slide_bg_warm.png` \| `light` \|
67	\| Dark \| `slide_bg_dark.png` \| `dark` \|
68
69	Do NOT ask for slide count. Generate as many slides as the content needs — images drive the
70	count, not a preset number.
71
72	---
73
74	## Phase 1: Research Context Discovery
75
76	Scan project docs (CLAUDE.md, README.md, memory files, `docs/`) to understand:
77	- Research question and direction
78	- Key metrics and what "good" vs "bad" looks like
79	- Terminology, model names, dataset names, output folder conventions
80	- Prior results and baselines
81	- Previous presentations (check `presentation/` directory for prior slide decks to recap)
82
83	If no research goal found, use `AskUserQuestion`:
84	- header: "Research context"
85	- question: "Brief description of what this project does?"
86	- options: ["Let me describe it", "Use README description"]
87
88	Use this context internally for smart organization (grouping, filtering, prioritization).
89	Do NOT let it leak into slide text.
90
91	---
92
93	## Phase 2: Experiment Discovery
94
95	Use ALL methods — each catches things the others miss.
96
97	### Git Log
98
99	```bash
100	git log --after="YYYY-MM-DD_START" --before="YYYY-MM-DD_END+1" --oneline --stat
101	```
102
103	Extract: commit messages (often contain results), files changed, dates.
104
105	### Output Folders
106
107	```bash
108	find . -maxdepth 2 -type d -newermt "YYYY-MM-DD_START" ! -newermt "YYYY-MM-DD_END+1" 2>/dev/null \| sort
109	```
110
111	Also check: `outputs/`, `results/`, `experiments/`, `runs/`, `logs/`, `checkpoints/`.
112	Do NOT assume MMDD_ naming — use modification time as primary signal.
113
114	For each folder: list contents, look for metrics files (`metrics.json`, `*.pkl`, `scores.txt`,
115	`.log`) and images (`.png`, `.jpg`, `.gif`).
116
117	### Scripts
118
119	```bash
120	find . -name "*.py" -newermt "YYYY-MM-DD_START" ! -newermt "YYYY-MM-DD_END+1" 2>/dev/null \| grep -v __pycache__ \| sort
121	```
122
123	Also check `.sh` scripts. Read the code to understand what each experiment does — folder
124	names are opaque, scripts tell you everything.
125
126	### Media Selection — AGGRESSIVE
127
128	Collect FAR more images than you think you need. This is the most important disc

People who liked this also viewed

6 picks

Type

results-to-slides

23installs0forks

View on GitHub

How to install results-to-slides?

saidwivedi/research-skills/results-to-slides

$npx -y skills add saidwivedi/research-skills --skill results-to-slides

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Results to Slides
2
3	Discover experiments, collect images/metrics, organize into slides. You provide the story —
4	this skill handles the grunt work. Output: slide markdown + editable PPTX.
5
6	## Core Rules
7
8	1. State what was done and what the result was. Do not editorialize. Never use:
9	"breakthrough", "key insight", "importantly". Just experiment + number.
10	2. Every slide MUST have a visual. No text-only slides (except pure tables). If a concept
11	has no experiment images, create a diagram. If no diagram makes sense, show dataset examples.
12	3. Images first, text second. Plan images for each slide BEFORE writing bullets. If you
13	can't find an image for a slide, restructure or merge it.
14
15	---
16
17	## Parallel Execution
18
19	Maximize use of the Agent tool. Whenever you have 2+ independent tasks, launch parallel agents.
20
21	- Phase 1: Launch agents in parallel to read CLAUDE.md, README.md, memory files
22	- Phase 2 (biggest win): Launch separate agents for git log, output folders, scripts, media
23	- Phase 3: Launch agents in parallel to read different experiment scripts and output folders
24	- Phase 5+6: Sequential (markdown must be written before converting)
25
26	---
27
28	## Important Paths
29
30	- Skill directory: `${CLAUDE_SKILL_DIR}`
31	- Converter: `${CLAUDE_SKILL_DIR}/md_to_pptx.py`
32	- Backgrounds: `${CLAUDE_SKILL_DIR}/backgrounds/`
33	- Theme CSS: `${CLAUDE_SKILL_DIR}/theme.css`
34	- Slide element reference: [slide_reference.md](slide_reference.md)
35
36	---
37
38	## Phase 0: Parse Arguments & Setup
39
40	The user provides: `/results-to-slides START_DATE END_DATE`
41
42	Arguments come as `$ARGUMENTS` containing two MMDD date strings (e.g., `0301 0308`).
43
44	Parse the dates:
45	- `$0` = start date (MMDD), `$1` = end date (MMDD)
46	- Infer year from system date. End date is INCLUSIVE (use end_date + 1 for `find` bounds)
47
48	If no arguments provided, use `AskUserQuestion`:
49	- header: "Date range"
50	- question: "What date range? Use MMDD MMDD format (e.g., 0301 0308)."
51	- options: ["Last week", "Last 2 weeks", "Last month"]
52
53	Output directory: `presentation/YYYY_MM_DD/` using the end date.
54
55	### Ask Presentation Preferences
56
57	Ask all preferences upfront before discovery, in a SINGLE AskUserQuestion call:
58
59	1. Background theme: "Light (Recommended)", "Warm", "Dark"
60	2. Emphasis areas: "What should the presentation emphasize? (e.g., dataset, model architecture,
61	results comparison, failed approaches)" — free text, helps prioritize slide count allocation.
62
63	\| Choice \| Background file \| `--theme` flag \|
64	\|--------\|----------------\|----------------\|
65	\| Light \| `slide_bg_light.png` \| `light` \|
66	\| Warm \| `slide_bg_warm.png` \| `light` \|
67	\| Dark \| `slide_bg_dark.png` \| `dark` \|
68
69	Do NOT ask for slide count. Generate as many slides as the content needs — images drive the
70	count, not a preset number.
71
72	---
73
74	## Phase 1: Research Context Discovery
75
76	Scan project docs (CLAUDE.md, README.md, memory files, `docs/`) to understand:
77	- Research question and direction
78	- Key metrics and what "good" vs "bad" looks like
79	- Terminology, model names, dataset names, output folder conventions
80	- Prior results and baselines
81	- Previous presentations (check `presentation/` directory for prior slide decks to recap)
82
83	If no research goal found, use `AskUserQuestion`:
84	- header: "Research context"
85	- question: "Brief description of what this project does?"
86	- options: ["Let me describe it", "Use README description"]
87
88	Use this context internally for smart organization (grouping, filtering, prioritization).
89	Do NOT let it leak into slide text.
90
91	---
92
93	## Phase 2: Experiment Discovery
94
95	Use ALL methods — each catches things the others miss.
96
97	### Git Log
98
99	```bash
100	git log --after="YYYY-MM-DD_START" --before="YYYY-MM-DD_END+1" --oneline --stat
101	```
102
103	Extract: commit messages (often contain results), files changed, dates.
104
105	### Output Folders
106
107	```bash
108	find . -maxdepth 2 -type d -newermt "YYYY-MM-DD_START" ! -newermt "YYYY-MM-DD_END+1" 2>/dev/null \| sort
109	```
110
111	Also check: `outputs/`, `results/`, `experiments/`, `runs/`, `logs/`, `checkpoints/`.
112	Do NOT assume MMDD_ naming — use modification time as primary signal.
113
114	For each folder: list contents, look for metrics files (`metrics.json`, `*.pkl`, `scores.txt`,
115	`.log`) and images (`.png`, `.jpg`, `.gif`).
116
117	### Scripts
118
119	```bash
120	find . -name "*.py" -newermt "YYYY-MM-DD_START" ! -newermt "YYYY-MM-DD_END+1" 2>/dev/null \| grep -v __pycache__ \| sort
121	```
122
123	Also check `.sh` scripts. Read the code to understand what each experiment does — folder
124	names are opaque, scripts tell you everything.
125
126	### Media Selection — AGGRESSIVE
127
128	Collect FAR more images than you think you need. This is the most important disc

People who liked this also viewed

6 picks

Type