✓
Purpose & Capability
Name, description, and the SKILL.md workflow are coherent: a market-risk co‑pilot that summarizes technical context, headlines, and writes a log. No unrelated env vars, binaries, or installs are requested.
!
Instruction Scope
The instructions are largely scoped to the stated task, but they are open-ended about obtaining 'recent news' and 'publicly available info' (leaving the agent discretion to fetch web content). More importantly, pre-scan detected unicode control characters in SKILL.md — this is a common prompt-injection technique used to hide or alter instructions and could change runtime behavior. Inspect the raw file before trusting autonomous runs.
✓
Install Mechanism
Instruction-only skill with no install spec or code files — lowest install risk; nothing will be written to disk by an installer.
✓
Credentials
No environment variables, credentials, or configuration paths are requested. Requested capabilities are proportional to a news-and-price summarization skill.
✓
Persistence & Privilege
always:false and no special privileges or config modifications requested. Autonomous invocation is permitted (platform default) but not combined with broad access.
Scan Findings in Context
[unicode-control-chars] unexpected: Hidden control characters in SKILL.md are not expected for a market-summary instruction file; they can be used to inject or obfuscate instructions that change model behavior. No other regex findings (no code files) were available for context.
What to consider before installing
What to consider before installing:
- The skill itself is coherent for market-headline scanning and logging and asks for no credentials or installs. That part looks benign.
- However, the SKILL.md contains hidden unicode control characters (a prompt-injection indicator). These characters can conceal or alter instructions the model will follow. Before installing, inspect the raw SKILL.md for non-printable/control characters (e.g., open in a hex editor, run `cat -v SKILL.md`, or use a tool that shows Unicode code points) and remove them if present.
- Because the workflow is open-ended about obtaining headlines, decide and restrict whether the agent is allowed to make live web requests. If you don't want network fetches, restrict the agent or require the user to paste sources.
- Do not provide any credentials or API keys to this skill. Test it first with non-sensitive, mock watchlists and verify that all cited headlines and timestamps are real and come from expected public sources.
- If possible, ask the publisher for a source/homepage or request a cleaned SKILL.md without hidden characters. If you can't verify the file's origin, treat installation as higher risk.
Confidence is medium because the skill is otherwise coherent, but the prompt-injection signal is a notable red flag that requires manual review before trusting autonomous use.
