The Agent Skills Directory

[COMMAND_EXECUTION]: Executes shell commands such as claude mcp add and export to configure the dkod MCP server and authentication tokens. These are used for initial setup and environment configuration.
[EXTERNAL_DOWNLOADS]: Configures and interacts with the dkod MCP server hosted at https://api.dkod.io/mcp. This external resource is owned by the skill vendor and is required for the parallel execution features.
[PROMPT_INJECTION]: The skill processes untrusted codebase data through tools like dk_file_read and dk_context, and has the capability to write changes back via dk_file_write and dk_push.
Ingestion points: dk_file_read and dk_context tools (documented in SKILL.md and references/mcp-workflow.md).
Boundary markers: Absent; no explicit instruction delimiters are mentioned for processing external code.
Capability inventory: dk_file_write, dk_merge, dk_push, and dk_submit allow for codebase modification and PR creation (documented in SKILL.md and references/mcp-workflow.md).
Sanitization: Absent; the skill performs AST-level structural analysis which does not validate against natural language instructions embedded in code comments or strings.

dkod