The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The script bin/topup-test-wallets.sh utilizes sudo -n -u postgres to execute database commands as the superuser. This allows for horizontal privilege escalation within the environment where the script is executed.
[COMMAND_EXECUTION]: The skill package includes administrative logic for remote operations. The script bin/topup-test-wallets.sh performs an ssh colo2 to execute SQL commands against a PostgreSQL instance. This indicates the skill is capable of performing high-risk remote command execution and infrastructure modification.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and process data from diverse, untrusted public sources including Reddit, Hacker News, and the Twitter archive.
Ingestion points: Data enters the agent's context through the /v1/scry/query and /v1/scry/embed endpoints, which return rows from public corpora.
Boundary markers: The skill notes that results are returned as plain JSON arrays in column order, but lacks explicit instructions for sanitizing these results before the LLM processes them.
Capability inventory: The skill has capabilities to write to a persistent server-side memory (memory_write MCP tool), perform network operations via curl, and influence future research directions through the creativity/outsized fan-out endpoint.
Sanitization: There is no mention of output sanitization or filtering for malicious instructions embedded in the public data (e.g., hidden markdown or code comments).
[CREDENTIALS_UNSAFE]: The bin/topup-test-wallets.sh script exposes internal administrative metadata, including specific User UUIDs (55fbd76f-4113-4a67-aebd-c4cc082f1127, 2b643878-5136-45d7-bb88-4a1f6b74578f) and references to internal secret storage paths in Vault (secret/secret/scry/test-account-pricing-lanes). While these are not plaintext passwords, they provide actionable intelligence for targeting the vendor's internal wallet systems.

scry