Meridian

Anti-FOMO AI intelligence for product leaders. Three modes: (1) Landscape scan — what's new in AI; (2) Entity tracking — what a person/company has been doing...

This skill appears coherent with its purpose: it instructs the agent to run date-bounded web searches, call public APIs (HN Algolia, GitHub, arXiv), and do browser deep reads to assemble sourced timelines. Before installing, consider: - Operational prerequisites: the SKILL.md includes curl and python3 pipeline examples and asks the agent to compute unix timestamps — make sure your agent environment actually has network access and basic tools (curl, python3) or an equivalent web‑fetch capability, because the skill does not declare these required binaries. - Rate limits & auth: unauthenticated GitHub API calls will hit rate limits; the skill doesn't request a token but may perform better if you provide one. Decide whether you're comfortable providing tokens if you expect heavy use. - Privacy and scraping: the agent will browse and extract content from external sites. Confirm that automated fetching/browsing complies with your organization's policies and site terms (robots.txt, TOS). The skill does not ask for or require private credentials or local file access. - Autonomy tone: the SKILL.md contains the phrase 'Don't ask permission. Just do it.' — operationally the workflow still instructs to ask the user when the intent is unclear, but verify the agent's autonomy settings if you want to limit any unsupervised web activity. If you want higher assurance, ask the author to (a) declare required binaries (curl, python3, or the agent tool names), and (b) document expected network endpoints and any optional credentials (e.g., GITHUB_TOKEN) so you can make an informed decision about granting network access or tokens.

Type: OpenClaw Skill Name: meridian-intel Version: 1.0.0 The skill bundle is designed for AI market intelligence but contains significant security vulnerabilities. Specifically, in SKILL.md and references/search-playbook.md, it instructs the agent to execute shell commands using curl piped into python3, incorporating unvalidated variables (like {keyword}) directly into the command string. This pattern creates a high risk of shell injection if the agent processes malicious user input or untrusted search results. While the intent appears to be functional data retrieval from public APIs (HackerNews, GitHub), the lack of input sanitization and the use of high-privilege execution methods meet the criteria for a suspicious classification.