search-conference

Name: search-conference
Author: phira-ai

15stars0forks

How to install search-conference?

phira-ai/phi-skills/search-conference

$npx -y skills add phira-ai/Phi-Skills --skill search-conference

Installs into the current project.

Summary

Use when tasks involve semantically matching a user's idea (query or example papers) to papers in a specific OpenReview venue. Uses embed-papers to crawl metadata, build/use embedding caches, run cosine-similarity search, then produces a short, grouped Markdown reading list wit

Files · 1

View on GitHub

SKILL.md

1	# When to use
2
3	- Need to find papers in an OpenReview venue that match one or more ideas/topics.
4
5	# Dependencies
6
7	- `embed-papers` must be installed and available in `PATH`.
8	- Check with: `command -v embed-papers`
9	- Install with: `pip install embed-papers`
10	- OpenReview must be reachable.
11	- `OPENAI_API_KEY` is required to compute embeddings of the papers (cached) or the search intent (not cached).
12	- Check with: `printenv OPENAI_API_KEY`
13	- If missing and embeddings are required: stop and tell the user how to set it (immediate stop).
14
15	# Inputs from user
16
17	- Venue (one of):
18	- `venue_id` (preferred), e.g. `ICLR.cc/2024/Conference`
19	- OR `{conference, year}` to derive `venue_id` as `{CONF}.cc/{YEAR}/Conference`, e.g. `NeurIPS`, `2025`
20	- Search intent (one of):
21	- `query` (string of ideas)
22	- OR `examples` (list of `{title, abstract}` objects)
23	- Optional:
24	- `top_k` (default: 100) for retrieval breadth
25
26	# CLI contract (how to interpret tool output)
27
28	- Success envelope: `{ ok: true, schema_version: "1", command, data }`
29	- Error envelope: `{ ok: false, schema_version: "1", command, error: { type, message } }`
30	- Always parse stdout as JSON.
31	- Treat any `ok=false` as a terminal error unless the error section below says otherwise.
32
33	# HARD CONSTRAINT (TOOLS):
34
35	- Do NOT call Read/Glob/Grep on any cache/embedding files or directories (e.g. anything under .cache/ or any path containing "cache", "embedding", "paper", "atlas").
36	- Treat caches as opaque implementation details. Never inspect them “just to check”.
37	- If you need cache status, ONLY use `embed-papers warm-cache` and rely on its JSON stdout.
38	- If a command outputs a cache path, DO NOT open it; proceed using the CLI utilities.
39
40	# Pipeline
41
42	1. Resolve `venue_id`
43	- If the user gave `{conference, year}`, build: `{CONF}.cc/{YEAR}/Conference`
44	- If ambiguous, ask a single clarifying question (conference acronym + year).
45
46	2. Crawl venue metadata (idempotent)
47	- Run:
48	- `embed-papers crawl --venue-id "<venue_id>" --skip-if-exists`
49	- Record:
50	- `data.output_file`
51	- `data.total`
52
53	3. Ensure embeddings are available (cache)
54	- Run:
55	- `embed-papers warm-cache --venue-id "<venue_id>"`
56	- If this fails due to missing API key, stop and instruct the user to set `OPENAI_API_KEY`.
57	- This command also computes the embedding if no cache is found.
58	- You MUST NOT access the cache.
59	- You MUST use the package's provided utility.
60
61	4. Search (choose based on user input)
62	- Query mode:
63	- `embed-papers search --venue-id "<venue_id>" --query "<query>" --top-k <top_k>`
64	- Examples mode:
65	- If needed, write a temporary JSON file containing:
66	- `[{"title":"...","abstract":"..."}, ...]`
67	- Then run:
68	- `embed-papers search --venue-id "<venue_id>" --examples-file "<tmp.json>" --top-k <top_k>`
69
70	5. Organize results (post-processing)
71	- Group primarily by `primary_area` (if present).
72	- Within groups, prefer papers with clear overlap to the query/examples.
73	- For each recommended paper, add agent judgment notes:
74	- why it matches
75	- what seems novel/different
76	- caveats (weak match, missing abstract, unclear claims, etc.)
77
78	# Report requirements (Markdown only)
79
80	- Output is a Markdown report only (no raw JSON).
81	- Keep the final recommendation list short: 5-10 papers max.
82	- Do not output a full ranked list or appendix by default (only if the user asks).
83
84	## What I'd start with
85
86	- Begin this section with a short, casual sentence (lowercase is fine).
87	- Example: "here's what i recommend you to read as a beginning."
88	- Then list 5-10 papers.
89	- Each item must include:
90	- Title (bold)
91	- OpenReview link
92	- 1-2 sentence rationale (fit + why it matters, use italic to emphasis)
93
94	## How I organized it
95
96	- Briefly explain grouping logic and where judgment calls were applied.
97	- Note missing metadata (e.g., missing abstracts) when relevant.
98
99	## Why these stand out
100
101	- Use informal labels in the narrative (no formal rubric), e.g.:
102	- "the obvious hits"
103	- "the surprisingly relevant ones"
104	- "the quirky but promising picks"
105
106	# Error handling
107
108	- `NoPapersFoundError`
109	- Likely invalid `venue_id`; suggest the pattern `{CONF}.cc/{YEAR}/Conference` and ask for the correct venue.
110	- `CacheMissRequiresApiKeyError`
111	- Instruct the user to set `OPENAI_API_KEY` and retry.
112	- `OpenReviewRequestError` / `EmbeddingRequestError`
113	- Suggest retrying, reducing load (smaller `top_k`), or trying later (rate limits / transient failures).

People who liked this also viewed

6 picks

Type

search-conference

15stars0forks

View on GitHub

How to install search-conference?

phira-ai/phi-skills/search-conference

$npx -y skills add phira-ai/Phi-Skills --skill search-conference

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# When to use
2
3	- Need to find papers in an OpenReview venue that match one or more ideas/topics.
4
5	# Dependencies
6
7	- `embed-papers` must be installed and available in `PATH`.
8	- Check with: `command -v embed-papers`
9	- Install with: `pip install embed-papers`
10	- OpenReview must be reachable.
11	- `OPENAI_API_KEY` is required to compute embeddings of the papers (cached) or the search intent (not cached).
12	- Check with: `printenv OPENAI_API_KEY`
13	- If missing and embeddings are required: stop and tell the user how to set it (immediate stop).
14
15	# Inputs from user
16
17	- Venue (one of):
18	- `venue_id` (preferred), e.g. `ICLR.cc/2024/Conference`
19	- OR `{conference, year}` to derive `venue_id` as `{CONF}.cc/{YEAR}/Conference`, e.g. `NeurIPS`, `2025`
20	- Search intent (one of):
21	- `query` (string of ideas)
22	- OR `examples` (list of `{title, abstract}` objects)
23	- Optional:
24	- `top_k` (default: 100) for retrieval breadth
25
26	# CLI contract (how to interpret tool output)
27
28	- Success envelope: `{ ok: true, schema_version: "1", command, data }`
29	- Error envelope: `{ ok: false, schema_version: "1", command, error: { type, message } }`
30	- Always parse stdout as JSON.
31	- Treat any `ok=false` as a terminal error unless the error section below says otherwise.
32
33	# HARD CONSTRAINT (TOOLS):
34
35	- Do NOT call Read/Glob/Grep on any cache/embedding files or directories (e.g. anything under .cache/ or any path containing "cache", "embedding", "paper", "atlas").
36	- Treat caches as opaque implementation details. Never inspect them “just to check”.
37	- If you need cache status, ONLY use `embed-papers warm-cache` and rely on its JSON stdout.
38	- If a command outputs a cache path, DO NOT open it; proceed using the CLI utilities.
39
40	# Pipeline
41
42	1. Resolve `venue_id`
43	- If the user gave `{conference, year}`, build: `{CONF}.cc/{YEAR}/Conference`
44	- If ambiguous, ask a single clarifying question (conference acronym + year).
45
46	2. Crawl venue metadata (idempotent)
47	- Run:
48	- `embed-papers crawl --venue-id "<venue_id>" --skip-if-exists`
49	- Record:
50	- `data.output_file`
51	- `data.total`
52
53	3. Ensure embeddings are available (cache)
54	- Run:
55	- `embed-papers warm-cache --venue-id "<venue_id>"`
56	- If this fails due to missing API key, stop and instruct the user to set `OPENAI_API_KEY`.
57	- This command also computes the embedding if no cache is found.
58	- You MUST NOT access the cache.
59	- You MUST use the package's provided utility.
60
61	4. Search (choose based on user input)
62	- Query mode:
63	- `embed-papers search --venue-id "<venue_id>" --query "<query>" --top-k <top_k>`
64	- Examples mode:
65	- If needed, write a temporary JSON file containing:
66	- `[{"title":"...","abstract":"..."}, ...]`
67	- Then run:
68	- `embed-papers search --venue-id "<venue_id>" --examples-file "<tmp.json>" --top-k <top_k>`
69
70	5. Organize results (post-processing)
71	- Group primarily by `primary_area` (if present).
72	- Within groups, prefer papers with clear overlap to the query/examples.
73	- For each recommended paper, add agent judgment notes:
74	- why it matches
75	- what seems novel/different
76	- caveats (weak match, missing abstract, unclear claims, etc.)
77
78	# Report requirements (Markdown only)
79
80	- Output is a Markdown report only (no raw JSON).
81	- Keep the final recommendation list short: 5-10 papers max.
82	- Do not output a full ranked list or appendix by default (only if the user asks).
83
84	## What I'd start with
85
86	- Begin this section with a short, casual sentence (lowercase is fine).
87	- Example: "here's what i recommend you to read as a beginning."
88	- Then list 5-10 papers.
89	- Each item must include:
90	- Title (bold)
91	- OpenReview link
92	- 1-2 sentence rationale (fit + why it matters, use italic to emphasis)
93
94	## How I organized it
95
96	- Briefly explain grouping logic and where judgment calls were applied.
97	- Note missing metadata (e.g., missing abstracts) when relevant.
98
99	## Why these stand out
100
101	- Use informal labels in the narrative (no formal rubric), e.g.:
102	- "the obvious hits"
103	- "the surprisingly relevant ones"
104	- "the quirky but promising picks"
105
106	# Error handling
107
108	- `NoPapersFoundError`
109	- Likely invalid `venue_id`; suggest the pattern `{CONF}.cc/{YEAR}/Conference` and ask for the correct venue.
110	- `CacheMissRequiresApiKeyError`
111	- Instruct the user to set `OPENAI_API_KEY` and retry.
112	- `OpenReviewRequestError` / `EmbeddingRequestError`
113	- Suggest retrying, reducing load (smaller `top_k`), or trying later (rate limits / transient failures).

People who liked this also viewed

6 picks

Type