ctf-similar-search helps an agent find writeups of challenges similar to the one it is working on. It wraps the Tavily search and extract APIs in a small shell and Python helper and gives rules for writing queries that return CTF writeups.
Queries for competition and challenge names are limited to three words, while code snippets and unique strings are exempt. The agent reviews the results first and extracts only the pages that look relevant.
What You Get
- A search.sh and search.py helper with search and extract commands for the Tavily API
- Query rules for competition and challenge names, bare challenge names and unique strings
- A four-step workflow: search, review results, extract selectively, then analyze and pivot
- Guidance on when to switch to category skills, OSINT or the writeup skill
Common Use Cases
- Finding the original writeup of a past competition challenge that resembles the current one
- Identifying a challenge type when only a name or a code fragment is known
- Looking up how others solved a challenge after several failed attempts
What Makes It Unique
It is the only skill in the collection that reaches outside the repository for reference material. The helper enforces the three-word limit on non-code queries in code, which keeps results on CTF writeups and away from broad vulnerability explainers, and it prints the credits each call used.
Under The Hood
The skill runs a shell script that calls a Python script. The Python script reads a Tavily API key from an environment variable and sends the search query or the URLs to the Tavily search and extract endpoints over the network with the requests package, then prints the results. The user supplies the key and installs the requests and tavily packages. The query text and URLs leave the machine for that third-party service.