The Agent Skills Directory

[COMMAND_EXECUTION]: Invokes the opensquilla CLI tool through an absolute Python interpreter path (/abs/path/python -P -m opensquilla.cli.main) to perform repository management and code editing tasks.
[REMOTE_CODE_EXECUTION]: Clones code from remote Git repositories and executes their native toolchains (such as npm ci, npm run build, and pytest) directly on the host system. The skill explicitly notes the lack of a sandbox environment for these operations.
[EXTERNAL_DOWNLOADS]: Fetches source code from remote Git URLs and retrieves issue descriptions and commentary through the GitHub CLI (gh).
[PROMPT_INJECTION]: Exposed to indirect prompt injection via the ingestion of untrusted data from external codebases and GitHub issue threads.
Ingestion points: Data is brought into the environment from Git repositories, GitHub issues, and issue comments.
Boundary markers: None identified; there are no clear delimiters or instructions provided to the agent to treat external data as untrusted.
Capability inventory: The skill possesses exec_command, background_process, and file system write capabilities.
Sanitization: The skill does not implement sanitization or filtering for the data ingested from external sources.

code-task