The Agent Skills Directory

[METADATA_POISONING]: The skill metadata and documentation direct users to websites identified as phishing risks by automated security analysis tools.
Evidence: bazaar.playhunch.xyz, www.playhunch.xyz, and https://bazaar.playhunch.xyz are flagged as malicious phishing domains by the URLite scanner.
[EXTERNAL_DOWNLOADS]: The skill instructions require the user to perform an external installation from a personal, unverified GitHub repository (rajkaria/hunch-skills) rather than a trusted organization or vendor.
Evidence: Reference to https://github.com/rajkaria/hunch-skills/tree/78803b097712d74657e2798e9fc0c68f6a3bd1c0/hunch-bazaar in SKILL.md and catalog.json.
[DATA_EXFILTRATION]: The skill facilitates the transfer of financial assets (USDC) to a centrally controlled settlement account instead of a decentralized escrow contract. Given the associated phishing alerts for the target domain, this model presents a high risk of asset loss.
Evidence: The skill explicitly states: "Payments go to the pinned Hunch-controlled settlement account, not a rule-enforcing market escrow contract."
[INDIRECT_PROMPT_INJECTION]: The skill allows the creation of markets based on arbitrary user input, creating a surface for prompt injection attacks that could influence agent behavior during market drafting or betting.
Ingestion points: User-provided phrases for market creation in SKILL.md (e.g., "make a market from a phrase").
Boundary markers: Absent; there are no instructions to delimit or ignore embedded instructions within the market phrase.
Capability inventory: The skill utilizes native sign_data and accesses permanent wallet files, as noted in catalog.json.
Sanitization: Absent; no validation, escaping, or filtering of the user-provided market descriptions is described.

hunch-bazaar