outline-agent

Name: outline-agent
Author: woodfishhhh

31stars1forks

How to install outline-agent?

woodfishhhh/ez_math_model/outline-agent

$npx -y skills add woodfishhhh/EZ_math_model --skill outline-agent

Installs into the current project.

Summary

Step 1 of the PaperOrchestra pipeline (arXiv:2604.05018). Convert (idea.md, experimental_log.md, template.tex, conference_guidelines.md) into a strict JSON outline containing a plotting plan, literature search plan (Intro + Related Work), and section-level writing plan with citat

Files · 1

View on GitHub

SKILL.md

1	# Outline Agent (Step 1)
2
3	Faithful implementation of the Outline Agent from PaperOrchestra
4	(Song et al., 2026, arXiv:2604.05018, App. F.1, pp. 40–44).
5
6	Cost: 1 LLM call.
7
8	## Your task
9
10	Read four input files from the workspace and produce a single JSON object at
11	`workspace/outline.json` with three top-level keys:
12
13	- `plotting_plan` — array of figure objects
14	- `intro_related_work_plan` — object with `introduction_strategy` and `related_work_strategy`
15	- `section_plan` — array of section objects, each with `section_title` and `subsections[]`
16
17	## How to do it
18
19	1. Read the verbatim prompt at `references/prompt.md`. This is the exact
20	Outline Agent system prompt from the paper. Use it as your system message.
21	2. Prepend the Anti-Leakage Prompt from
22	`../paper-orchestra/references/anti-leakage-prompt.md`.
23	3. Read the four input files:
24	- `workspace/inputs/idea.md`
25	- `workspace/inputs/experimental_log.md`
26	- `workspace/inputs/template.tex`
27	- `workspace/inputs/conference_guidelines.md`
28	4. Synthesize across all four — the global instruction in the prompt is
29	"Do not analyze inputs in isolation. You must synthesize information across
30	all provided documents for every step."
31	5. Emit a single JSON object following the schema in
32	`references/outline-schema.md`. Cross-check against
33	`references/outline_schema.json` (machine-readable).
34	6. Save to `workspace/outline.json`.
35	7. Validate:
36	```bash
37	python skills/outline-agent/scripts/validate_outline.py workspace/outline.json
38	```
39	If validation fails, fix the JSON and re-validate. Do not proceed to Step 2
40	or Step 3 with an invalid outline — every downstream agent depends on this
41	schema.
42
43	## Hard rules from the prompt (do not violate)
44
45	These are excerpted from `references/prompt.md`. The validator enforces them.
46
47	### Plotting plan (Directive 1)
48
49	- `plot_type` MUST be exactly one of `"plot"` or `"diagram"`.
50	- `data_source` MUST be exactly one of `"idea.md"`, `"experimental_log.md"`,
51	or `"both"`.
52	- `aspect_ratio` MUST be exactly one of:
53	`"1:1"`, `"1:4"`, `"2:3"`, `"3:2"`, `"3:4"`, `"4:1"`, `"4:3"`, `"4:5"`,
54	`"5:4"`, `"9:16"`, `"16:9"`, `"21:9"`.
55	- `figure_id` MUST be a semantically meaningful snake_case identifier
56	(e.g., `fig_framework_overview`, `fig_ablation_study_parameter_sensitivity`).
57	- `figure_id` MUST NOT contain the word `"Figure"`.
58
59	### Intro / Related Work strategy (Directive 2)
60
61	- Strictly separate Introduction (macro-level context, 10-20 papers,
62	foundational + survey + impact) from Related Work (micro-level technical
63	baselines, 30-50 papers, divided into 2-4 methodology clusters that
64	directly compete with or precede the proposed approach).
65	- For each Related Work cluster: provide `methodology_cluster`,
66	`sota_investigation_mission`, `limitation_hypothesis`,
67	`limitation_search_queries`, `bridge_to_our_method`.
68	- CRITICAL TIMELINE RULE: Do not instruct searches for any papers
69	published after `{cutoff_date}`. Derive `cutoff_date` from
70	`conference_guidelines.md` (e.g., "ICLR 2025 → cutoff October 2024",
71	"CVPR 2025 → cutoff November 2024"). If unspecified, default to one month
72	before today's date.
73
74	### Section plan (Directive 3)
75
76	- Structural hierarchy: if Subsection X.1 is created, X.2 is mandatory.
77	No orphaned subsections. Omit subsections entirely if a section does not
78	require division.
79	- Content specificity: each `content_bullets` entry must reference source
80	materials concretely. AVOID "Describe the model". REQUIRE "Formalize the
81	Temporal-Aware Attention mechanism using Eq. 3 from idea.md."
82	- Mandatory citations: every dataset, optimizer, metric, and
83	foundational architecture/model mentioned in `idea.md` or
84	`experimental_log.md` MUST have a citation hint, no matter how ubiquitous
85	(e.g., AdamW, ResNet, ImageNet, CLIP, Transformer, LLaMA, GPT, LLaVA).
86	- Citation hint format:
87	- If you know the exact author and title:
88	`"Author (Exact Paper Title)"`
89	- Otherwise: `"research paper or technical report introducing '[Exact Model/Dataset/Metric Name]'"`
90	- Do NOT guess or hallucinate authors.
91
92	## Output
93
94	Exactly one file: `workspace/outline.json`. No prose, no code blocks, no
95	markdown. The Section Writing Agent and Literature Review Agent will parse
96	this JSON directly.
97
98	See `references/example-output.json` for a complete worked example from the
99	paper (App. F.1, pp. 43–44).
100
101	## Resources
102
103	- `references/prompt.md` — verbatim Outline Agent prompt from App. F.1
104	- `references/out

People who liked this also viewed

6 picks

Type

outline-agent

31stars1forks

View on GitHub

How to install outline-agent?

woodfishhhh/ez_math_model/outline-agent

$npx -y skills add woodfishhhh/EZ_math_model --skill outline-agent

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Outline Agent (Step 1)
2
3	Faithful implementation of the Outline Agent from PaperOrchestra
4	(Song et al., 2026, arXiv:2604.05018, App. F.1, pp. 40–44).
5
6	Cost: 1 LLM call.
7
8	## Your task
9
10	Read four input files from the workspace and produce a single JSON object at
11	`workspace/outline.json` with three top-level keys:
12
13	- `plotting_plan` — array of figure objects
14	- `intro_related_work_plan` — object with `introduction_strategy` and `related_work_strategy`
15	- `section_plan` — array of section objects, each with `section_title` and `subsections[]`
16
17	## How to do it
18
19	1. Read the verbatim prompt at `references/prompt.md`. This is the exact
20	Outline Agent system prompt from the paper. Use it as your system message.
21	2. Prepend the Anti-Leakage Prompt from
22	`../paper-orchestra/references/anti-leakage-prompt.md`.
23	3. Read the four input files:
24	- `workspace/inputs/idea.md`
25	- `workspace/inputs/experimental_log.md`
26	- `workspace/inputs/template.tex`
27	- `workspace/inputs/conference_guidelines.md`
28	4. Synthesize across all four — the global instruction in the prompt is
29	"Do not analyze inputs in isolation. You must synthesize information across
30	all provided documents for every step."
31	5. Emit a single JSON object following the schema in
32	`references/outline-schema.md`. Cross-check against
33	`references/outline_schema.json` (machine-readable).
34	6. Save to `workspace/outline.json`.
35	7. Validate:
36	```bash
37	python skills/outline-agent/scripts/validate_outline.py workspace/outline.json
38	```
39	If validation fails, fix the JSON and re-validate. Do not proceed to Step 2
40	or Step 3 with an invalid outline — every downstream agent depends on this
41	schema.
42
43	## Hard rules from the prompt (do not violate)
44
45	These are excerpted from `references/prompt.md`. The validator enforces them.
46
47	### Plotting plan (Directive 1)
48
49	- `plot_type` MUST be exactly one of `"plot"` or `"diagram"`.
50	- `data_source` MUST be exactly one of `"idea.md"`, `"experimental_log.md"`,
51	or `"both"`.
52	- `aspect_ratio` MUST be exactly one of:
53	`"1:1"`, `"1:4"`, `"2:3"`, `"3:2"`, `"3:4"`, `"4:1"`, `"4:3"`, `"4:5"`,
54	`"5:4"`, `"9:16"`, `"16:9"`, `"21:9"`.
55	- `figure_id` MUST be a semantically meaningful snake_case identifier
56	(e.g., `fig_framework_overview`, `fig_ablation_study_parameter_sensitivity`).
57	- `figure_id` MUST NOT contain the word `"Figure"`.
58
59	### Intro / Related Work strategy (Directive 2)
60
61	- Strictly separate Introduction (macro-level context, 10-20 papers,
62	foundational + survey + impact) from Related Work (micro-level technical
63	baselines, 30-50 papers, divided into 2-4 methodology clusters that
64	directly compete with or precede the proposed approach).
65	- For each Related Work cluster: provide `methodology_cluster`,
66	`sota_investigation_mission`, `limitation_hypothesis`,
67	`limitation_search_queries`, `bridge_to_our_method`.
68	- CRITICAL TIMELINE RULE: Do not instruct searches for any papers
69	published after `{cutoff_date}`. Derive `cutoff_date` from
70	`conference_guidelines.md` (e.g., "ICLR 2025 → cutoff October 2024",
71	"CVPR 2025 → cutoff November 2024"). If unspecified, default to one month
72	before today's date.
73
74	### Section plan (Directive 3)
75
76	- Structural hierarchy: if Subsection X.1 is created, X.2 is mandatory.
77	No orphaned subsections. Omit subsections entirely if a section does not
78	require division.
79	- Content specificity: each `content_bullets` entry must reference source
80	materials concretely. AVOID "Describe the model". REQUIRE "Formalize the
81	Temporal-Aware Attention mechanism using Eq. 3 from idea.md."
82	- Mandatory citations: every dataset, optimizer, metric, and
83	foundational architecture/model mentioned in `idea.md` or
84	`experimental_log.md` MUST have a citation hint, no matter how ubiquitous
85	(e.g., AdamW, ResNet, ImageNet, CLIP, Transformer, LLaMA, GPT, LLaVA).
86	- Citation hint format:
87	- If you know the exact author and title:
88	`"Author (Exact Paper Title)"`
89	- Otherwise: `"research paper or technical report introducing '[Exact Model/Dataset/Metric Name]'"`
90	- Do NOT guess or hallucinate authors.
91
92	## Output
93
94	Exactly one file: `workspace/outline.json`. No prose, no code blocks, no
95	markdown. The Section Writing Agent and Literature Review Agent will parse
96	this JSON directly.
97
98	See `references/example-output.json` for a complete worked example from the
99	paper (App. F.1, pp. 43–44).
100
101	## Resources
102
103	- `references/prompt.md` — verbatim Outline Agent prompt from App. F.1
104	- `references/out

People who liked this also viewed

6 picks

Type