The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with the local system by executing shell commands to perform wallet and network operations.
Evidence: The scripts scripts/eco_fund.py and scripts/routed_pay.py use subprocess.run to call the mm CLI for transaction signing and curl for making network requests.
[EXTERNAL_DOWNLOADS]: The skill retrieves external data and interacts with payment APIs to facilitate its purchasing functionality.
Evidence: scripts/discover.py fetches a service catalog from https://catalog.selat.ai. scripts/eco_fund.py communicates with https://api.eco.com and https://gateway-api.circle.com for gasless funding operations.
[DATA_EXFILTRATION]: The skill accesses wallet identifiers and maintains a persistent local record of financial transactions.
Evidence: Uses mm wallet list to obtain the user's public address and appends detailed purchase logs (merchant, amount, receipt) to a local file at ~/.selat-purchasing/ledger.jsonl.
[PROMPT_INJECTION]: The skill processes untrusted metadata from third-party merchants, which presents an indirect prompt injection attack surface.
Ingestion points: Merchant 402 challenge bodies and service catalog metadata are ingested to build requests and rank candidates.
Boundary markers: The documentation explicitly instructs the agent to trust schemas over examples, and scripts/x402ext.py provides a boundary for extensions.
Capability inventory: The skill has the ability to trigger transaction signatures via the mm CLI and perform network requests via curl and urllib.
Sanitization: scripts/x402ext.py implements a scrubbing mechanism to discard unverified reputation attestations (selatTransactabilityIndex) from untrusted origins.

selat-purchasing