The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates interaction with Electrobun applications through a local CLI tool and scripts.
It executes shell commands using the Bash tool, specifically targeting scripts/agent-electrobun.ts and related aliases.
Local state, including accessibility tree references and snapshots, is managed via JSON files in the /tmp directory.
[REMOTE_CODE_EXECUTION]: The skill provides the capability to evaluate arbitrary JavaScript within the target application's execution environment.
Commands such as agent-electrobun eval and agent-electrobun shell eval allow for direct script execution in webviews.
Support for Base64-encoded payloads is included to handle complex scripts, though no malicious encoded payloads were found in the skill's own documentation.
[PROMPT_INJECTION]: An indirect prompt injection surface is present as the skill processes untrusted data from external application UIs.
Ingestion points: Data enters the agent's context through snapshot (accessibility tree), get text, get html, get value, and get title commands.
Boundary markers: The instructions do not specify the use of delimiters or specific warnings to ignore instructions embedded in the processed application data.
Capability inventory: The skill possesses significant capabilities, including file system access (screenshots, repo paths) and arbitrary JavaScript evaluation.
Sanitization: There is no evidence of sanitization or filtering of the content read from the application before it is presented to the agent.

agent-electrobun