How to use the Communications Literature Review agent skill

What agent skills are and what this one makes the agent do

An agent skill is a set of instructions the agent loads when a request matches the skill's description. This skill's description covers communications, wireless, networking, satellite/NTN, Wi-Fi, cellular, transport protocols, congestion control, routing, scheduling, MAC/PHY, rate adaptation, channel estimation, beamforming and communication-system research where the user wants papers, related work, a survey or a landscape summary.

The skill is knowledge-base-first. It asks the agent to search your own sources before external ones: Zotero, then an Obsidian vault, then local PDF folders, and only then academic databases (IEEE Xplore, ScienceDirect, ACM Digital Library) and the broader web. It groups papers by technical theme rather than by search order and extracts details such as scenario and assumptions, core method, main result and limitation, labelling workshop papers and preprints.

Install the comm-lit-review skill

Download the ZIP from the skill's page on this directory. The package contains the skill file SKILL.md. Unzip it into one of two locations: your personal skills folder at ~/.claude/skills/, or a project's .claude/skills/ folder if you want the skill available only inside that project.

You can also upload the skill in the Claude apps through Settings → Capabilities → Skills, if your plan and app version offer that option.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into the project's .claude/skills/ folder
• App install: upload the ZIP under Settings → Capabilities → Skills

Trigger the skill with natural requests

You do not need a special command. Describe a communications research task in plain words and say what you want back, such as papers, related work, a survey or a landscape summary. The skill's description is matched against your request, so a clear topic plus a clear deliverable works best.

The skill also reads a source directive. Adding '— sources:' followed by source names limits the search to those sources. If you omit it, the default set is zotero, obsidian, local, ieee, sciencedirect, acm and web. The value 'all' means the full default set.

• Find related work on rate adaptation for Wi-Fi links and summarise the main approaches.
• Give me a survey of beamforming papers for LEO satellite systems, from my Zotero library first.
• Write a related-work section draft on congestion control for cellular networks. — sources: zotero, ieee
• Landscape summary of routing and scheduling for multi-hop wireless mesh networks, recent work only.

Choose sources, time windows and venues

If you do not give a year range, the skill asks for a short foundational set (before 2022) and a recent set (2022 to present). Preprints are labelled as preprint, workshop papers as workshop, and when a preprint and a formal version both exist the formal version is cited first.

Venues are ranked in tiers. Tier A includes journals such as IEEE JSAC, IEEE/ACM Transactions on Networking and IEEE Transactions on Wireless Communications, and conferences such as ACM SIGCOMM, USENIX NSDI and IEEE INFOCOM. The tiers are a soft priority. If you write 'only top venues', the skill treats Tier A as a hard filter.

If your Zotero, Obsidian or PDF sources are not connected or not present, the skill is written to skip them silently and continue.

• Local PDFs are read from papers/ and literature/ in the current project, or from a custom path set under '## Paper Library' in CLAUDE.md
• At most 20 local PDFs are scanned; if there are more, they are prioritised by filename and first-page relevance

Read the output and check it

The main output is a literature table with the columns Paper, Venue, Year, Layer, Scenario, Method, Key Result, Limitation, Relevance and Source, where Source shows where each paper came from first. After the table the skill summarises what the field solves, how papers cluster into two to four approaches, what you already had versus what was newly found, where evidence is strong or weak, and what gap remains. It ends with a Practical Takeaway section.

The skill's own guidance is that the agent may misread papers or miss sources, so verify each claim against the original publisher page before you cite it.

• Check the DOI or publisher page for every paper in the table
• Confirm assumptions and results in the paper itself, not only in the summary

Safety notes: network, scripts and data sharing

According to the listing, this skill does not run scripts but does use the network. Its allowed tools include Bash, Read, Glob, Grep, WebSearch, WebFetch, Write and the optional Zotero and Obsidian MCP tools. Searches of IEEE Xplore, ScienceDirect and the ACM Digital Library send your topic terms to those third-party sites, so do not use the skill if you cannot share your queries externally.

Full text on those databases may require institutional access. The skill definition itself does not control how the agent behaves; actual searching depends on the agent runtime and any connected tools.

• Do not use it for confidential or unpublished topic queries if external search is not acceptable
• Generic machine learning or software/API documentation questions are outside its scope

Tips for better results

Give the topic, the layer (PHY/MAC, transport, NTN or cross-layer) and the time window. The skill separates transport-layer rate control from PHY/MAC rate adaptation unless told otherwise, so naming the layer prevents mixing them up. Connecting Zotero or an Obsidian vault gives the skill your own annotated papers, which it treats as high-priority evidence.

Ask for the output in the stages you need. If the first pass is sparse, the skill widens from Tier A to Tier B and then Tier C venues before using the broader web.

• Name the layer: PHY/MAC, transport or NTN
• Put your existing papers in Zotero with tags or collections
• Ask for foundational and recent work separately if you need both