The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/run_dcode.py executes the dcode CLI and fallback runner scripts (e.g., run_claude.py) using subprocess.run. The command is passed as an argument list, which effectively prevents shell injection attacks.
[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection by reading workspace data via the --prompt-file and --session-file flags and interpolating it into the prompt sent to the dcode agent. * Ingestion points: File content is read from user-specified paths in scripts/run_dcode.py. * Boundary markers: There are no delimiters or system instructions used to isolate the ingested content from the rest of the prompt. * Capability inventory: The skill executes the dcode CLI, which may have write access to the workspace depending on the configured role. * Sanitization: The skill does not sanitize or escape the content of the ingested files.
[DYNAMIC_EXECUTION]: The skill dynamically modifies the sys.path to load modules from local shared directories and resolves the paths of fallback runner scripts at runtime. It also allows the path of the dcode executable to be customized via the DCODE_CLI_PATH environment variable.

dcode-runner