The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user to download and execute a shell script directly from a remote URL (https://github.com/midnightntwrk/compact/releases/latest/download/compact-installer.sh) using the curl | sh pattern. This allows for arbitrary code execution on the local system from an unverified external source.
[COMMAND_EXECUTION]: Provides instructions to run various commands that can execute arbitrary code, including npm install, npm run dev, and a custom asset synchronization script (scripts/sync-zk-assets.mjs) that performs file system operations like rmSync and cpSync on the project directory.
[EXTERNAL_DOWNLOADS]: The project configuration downloads numerous dependencies from the NPM registry and refers the user to a third-party domain (1am.dev) for browser extension installation and ZK proving assets.
[PROMPT_INJECTION]: The skill documents an architecture vulnerable to indirect prompt injection by processing untrusted data from an external indexer.
Ingestion points: The fetchLockerState function in lib/locker.ts reads data from an external indexerUri provided in the session configuration.
Boundary markers: There are no delimiters or instructions provided to the agent to ignore potentially malicious content embedded in the indexer's response.
Capability inventory: The skill possesses the capability to submit blockchain transactions and move funds through submitCallTxAsync and submitTxAsync in lib/locker.ts.
Sanitization: The decodeLockerState function performs direct deserialization of the hex-encoded data returned by the indexer into ledger state without validation or sanitization.

example-locker-dapp