github-featured/wanshuiyin--auto-claude-code-research-in-sleep

Files

T

Ruofeng Yang 5d6e40a66b docs: batch-fix drift — broken links, stale model refs, deprecation markers, name mismatch, PII categories

Health-audit batch (all items verified against the tree before fixing):

- docs/MODEL_COMBINATIONS{,_CN}.md: all 24 guide links were repo-root-relative
  inside docs/ (docs/X.md → resolves to docs/docs/X.md); stripped the prefix.
  docs/MiniMax-GLM-Configuration.md: 2 links pointed at nonexistent
  docs/README.md; now point at MODEL_COMBINATIONS_CN.md where the anchors live.
  skills/research-pipeline/SKILL.md: resumable-runs link missing ../.
- skills-codex-claude-review overlay: research-refine (5 spots) + paper-figure
  (1 spot) prose still said the reviewer is GPT-5.4 — in an overlay whose whole
  point is Claude-as-reviewer via claude-review MCP (the invocation blocks were
  already correct; only prose was stale). Now says "the Claude reviewer".
- docs/ARIS_INTRO.md: "GPT-5.5 / 5.5" blind-replacement leftovers (EN+CN spots);
  docs/ARIS_INTRO.html: 11 stale GPT-5.4 refs synced to the .md;
  hero SVGs (aris_logo, hero_idea, hero_interaction, hero_combined): reviewer
  label "GPT-5.4 xhigh" → "GPT-5.5 xhigh" (same char width, no layout shift).
- READMEs' community-skills rows: /paper-poster now marked
  "(deprecated → paper-poster-html)"; skills-codex-gemini-review/paper-slides
  still cited retired /paper-poster where mainline+mirror say /paper-poster-html.
- skills/comm-lit-review/SKILL.md: frontmatter name "comm-lit-review-claude-single"
  → aligned to directory name (old name referenced nowhere; mirror already used
  the dir name).
- Personal-info hygiene: interview-cheatsheet's personal-info banlist (mainline +
  codex mirror, 2 spots each) listed the owner's actual lab/degree/server strings
  — the banlist itself was the leak; now defines CATEGORIES only. A tutorial
  review.json re-disclosed the redacted strings in its fix note; genericized.
- README.md spotlight: Anti-Autoresearch counts updated to v0.5
  (46 integrity across families A–H + 13 AI-style impressions + 2 advisory = 61).

Known-remaining (deliberately out of scope): CN README lags EN by ~9 News
entries + the Anti-Autoresearch spotlight (translation batch); GPT-5.4 mentions
in kill-argument/resubmit-pipeline/slides-polish and dated News entries are
deliberate legacy options/history, untouched.

Verification: link checker 0 bad across the three docs; inventory consistent;
full suite 407 passed / 16 skipped.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

2026-07-02 16:14:00 +08:00

8.5 KiB

Raw Permalink Blame History

name, description, allowed-tools

name	description	allowed-tools
comm-lit-review	Communications-domain literature review with Claude-style knowledge-base-first retrieval. Use when the task is about communications, wireless, networking, satellite/NTN, Wi-Fi, cellular, transport protocols, congestion control, routing, scheduling, MAC/PHY, rate adaptation, channel estimation, beamforming, or communication-system research and the user wants papers, related work, a survey, or a landscape summary.	Bash(), Read, Glob, Grep, WebSearch, WebFetch, Write, mcp__zotero__, mcp__obsidian-vault__*

Comm Lit Review Claude Single

Research topic: $ARGUMENTS

Purpose

Use this skill for communications-domain literature review when the topic is about:

wireless communications
cellular systems, 4G/5G/6G, NR, NTN
satellite, LEO, GEO, integrated space-air-ground systems
Wi-Fi, WLAN, mesh, ad hoc, sidelink, V2X
routing, scheduling, resource allocation, beamforming
rate adaptation, link adaptation, ACM, HARQ, CSI feedback
transport protocols and congestion control in communication networks
cross-layer optimization for communication systems

If the center of gravity is generic ML architecture research, pure control theory without communications literature, or software/API documentation rather than papers, fall back to a general literature skill.

Constants

PAPER_LIBRARY: Check local PDFs in this order:
1. papers/ in the current project
2. literature/ in the current project
3. Custom path specified by the user in CLAUDE.md under ## Paper Library
MAX_LOCAL_PAPERS = 20: Maximum number of local PDFs to scan. If there are more, prioritize by filename and first-page relevance.

Source Selection

Parse $ARGUMENTS for a — sources: directive.

If — sources: is specified, only search the listed sources.
If not specified, default to:
- zotero
- obsidian
- local
- ieee
- sciencedirect
- acm
- web

Valid source values:

zotero
obsidian
local
ieee
sciencedirect
acm
web
all

If all is specified, interpret it as the full default source set.

Retrieval Order

This is a knowledge-base-first skill. Search in this order unless the user overrides it:

Zotero
Obsidian
local papers/ and literature/
IEEE Xplore
ScienceDirect
ACM Digital Library
broader web

Graceful degradation rules:

If a source is unavailable, do not fail.
Skip it silently.
Continue to the next source.

External Search Policy

For external search:

prefer IEEE Xplore first
then ScienceDirect
then ACM
then broader web only when needed

Publication policy:

prefer peer-reviewed journals and major conferences
label workshop papers as workshop
label arXiv-only or author-hosted versions as preprint
if both preprint and formal version exist, cite the formal version first

Time-window policy:

if the user does not specify a year range, include both a short foundational set and a recent set
recommended split:
- foundational: before 2022
- recent: 2022 to present

Venue Priority

Within each database tier, search venue tiers in this order.

Tier A

Journals:

IEEE Journal on Selected Areas in Communications (JSAC)
IEEE/ACM Transactions on Networking (ToN)
IEEE Transactions on Wireless Communications (TWC)
IEEE Transactions on Communications (TCOM)

Conferences:

ACM SIGCOMM
USENIX NSDI
ACM MobiCom
ACM CoNEXT
IEEE INFOCOM

Tier B

Journals:

IEEE Transactions on Vehicular Technology (TVT)
IEEE Wireless Communications Letters (WCL)
IEEE Communications Letters
Computer Networks
Computer Communications
Ad Hoc Networks
Physical Communication

Conferences:

IEEE ICC
IEEE GLOBECOM
IEEE WCNC
IEEE PIMRC
ACM MobiHoc

Tier C

other relevant IEEE journals and transactions
other relevant Elsevier journals
other clearly relevant ACM conferences and workshops
topic-specific satellite, optical, vehicular, IoT, aerial, or edge communications venues

Usage rules:

start from Tier A
widen to Tier B if needed
widen to Tier C if still sparse
only then broaden to full web search
by default this is a soft priority, not a hard whitelist
if the user says only top venues, top journals only, or top conferences only, treat Tier A as a hard filter

Workflow

Step 0a: Search Zotero Library

Skip this step if Zotero MCP is not configured or zotero is not enabled.

If available:

search by topic
capture title, authors, year, venue
pull user annotations, tags, or collections when present
treat these as high-priority evidence because they reflect the user's existing library

Step 0b: Search Obsidian Vault

Skip this step if Obsidian MCP is not configured or obsidian is not enabled.

If available:

search topic-related notes
collect summaries, wikilinks, tags, and paper references
treat these notes as the user's processed understanding of the topic

Step 0c: Scan Local Paper Library

Run this step if local is enabled.

locate PDFs from papers/**/*.pdf and literature/**/*.pdf
de-duplicate against Zotero hits when possible
read the first pages of relevant PDFs
extract title, authors, year, problem, method, and relevance
use local hits to guide and de-duplicate later external search

Step 1: Search External Primary Sources

Use a layered search strategy. For communications topics, avoid random blog posts or tertiary summaries.

Database ladder:

ieeexplore.ieee.org
sciencedirect.com
dl.acm.org
broader web using primary publisher pages, official conference sites, DOI pages, and author-hosted copies of already-identified formal papers

Move to the next database tier only when:

the higher-priority tier is too sparse
the topic clearly publishes elsewhere
the user explicitly asks for broader coverage

Within each database tier:

start from Tier A venues
widen to Tier B if needed
widen to Tier C if still sparse

Step 2: Extract Paper-Level Facts

For each relevant paper, capture:

Title
Authors
Year
Venue
Layer or system scope
Scenario and assumptions
Core method
Main result or claim
Limitation
Relevance to the user's topic
Source URL
Source origin: zotero, obsidian, local, ieee, sciencedirect, acm, or web

Favor concrete numbers, assumptions, and problem definitions over generic paraphrases.

Do not collapse transport-layer rate control and PHY/MAC rate adaptation into one bucket without saying so explicitly.

Synthesis Rules

Group papers by technical axis rather than by search order. Common groupings:

PHY/MAC adaptation
transport and congestion control
NTN and satellite resource management
cross-layer or learning-based control
measurement and empirical studies

When useful, explicitly separate:

foundational vs recent work
formal publications vs preprints
top-tier vs lower-tier venues
single-link vs multi-user formulations
simulation-only vs deployment-backed work
user-owned sources vs newly surfaced external papers

If evidence is weak, say so instead of smoothing it over.

Output

Use a literature table with these columns:

Paper	Venue	Year	Layer	Scenario	Method	Key Result	Limitation	Relevance	Source

Source should indicate where the paper came from first:

zotero
obsidian
local
ieee
sciencedirect
acm
web

After the table, summarize in this order:

what the field is mostly trying to solve
how papers cluster into 2-4 approaches
what the user already had vs what was newly surfaced
where the evidence is strong vs weak
what research gap remains

End with Practical Takeaway:

dominant current approach
likely saturated direction
promising open direction

Key Rules

Never fail because Zotero or Obsidian MCP is missing.
Prefer user-owned sources first when available, but do not let them replace external validation.
Prefer primary formal sources over summaries or tertiary commentary.
Prefer IEEE and ScienceDirect first, ACM second, and only then broader web search unless the user asks otherwise.
Search venue tiers from top to broad within each database tier.
Treat venue tiers as soft ranking by default and hard constraint only when the user explicitly asks for top-only search.
Do not pretend a preprint is peer reviewed.
If the topic spans multiple layers, say that the literature itself is split across layers.

8.5 KiB Raw Permalink Blame History