The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes the local devin CLI with the --permission-mode dangerous flag. This configuration explicitly instructs the agent to auto-approve all tool executions and command-line actions without manual user oversight, which is necessary for autonomy but increases the risk of destructive actions.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted user input via the prompt and messages fields and passes them to an autonomous agent. Ingestion points: Data enters through the run and session::create functions in SKILL.md. Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present. Capability inventory: The agent can execute local shell commands, perform network operations, and invoke other connected services via the iii trigger tool described in SKILL.md. Sanitization: There is no evidence of input validation or filtering.- [EXTERNAL_DOWNLOADS]: The skill relies on the external devin-cli tool, which the documentation recommends installing via Homebrew. While Devin is a recognized service, this introduces a dependency on external code that operates with high privileges.- [REMOTE_CODE_EXECUTION]: The skill facilitates remote code execution by design, as the Devin agent is capable of generating and running arbitrary code both locally and in its cloud environment based on the provided instructions.

devin