psbt
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONCREDENTIALS_UNSAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes base64-encoded PSBT strings provided as user input, which could potentially contain malicious structures or metadata.
- Ingestion points: The
psbtargument inestimate-fee,sign, andbroadcastsubcommands inpsbt.ts. - Capability inventory: The skill can sign transaction inputs using private keys (
tx.signIdx) and broadcast transactions to the Bitcoin network (mempool.broadcastTransaction). - Boundary markers: The
AGENT.mdfile contains logic to prevent signing index 0 (seller input), but there are no programmatic delimiters to isolate the PSBT content from the agent's control flow. - Sanitization: The skill uses the
@scure/btc-signerlibrary to decode and validate the PSBT structure, which provides a layer of protection against malformed data. - [COMMAND_EXECUTION]: The skill facilitates the broadcasting of transactions to the public Bitcoin network.
- The
broadcastcommand inpsbt.tstakes a finalized PSBT and submits it to the mempool.space API, resulting in an immutable change to the Bitcoin ledger. - [CREDENTIALS_UNSAFE]: The skill programmatically accesses sensitive Bitcoin private keys.
- Evidence: In
psbt.ts, thesigncommand retrieves the active account fromwalletManagerand accessesaccount.btcPrivateKeyandaccount.taprootPrivateKeyto perform signing operations.
Audit Metadata