The Agent Skills Directory

[PROMPT_INJECTION]: The skill utilizes external inputs such as ticket URLs or task descriptions via the WORK_ITEM and WORK_ITEM_LOOKUP fields. This represents an indirect prompt injection surface. The risk is mitigated by explicit instructions to ask the user for confirmation, follow established repository conventions, and wait for human review of 'Atomic commit groups' before applying changes.
Ingestion points: WORK_ITEM (ticket/task description), WORK_ITEM_LOOKUP (shard path/URL).
Boundary markers: Includes an intake phase where the agent must ask the user for missing values and wait for a response.
Capability inventory: Local file system writes, branch creation, and execution of validation commands.
Sanitization: Relies on LLM guardrails and the structured 'Step 1: Setup and Plan' process which requires approval before execution.
[COMMAND_EXECUTION]: The skill facilitates standard development operations, such as running mdcp check and executing repository-specific test suites as part of the TDD and validation phases. These actions are within the expected scope for a software engineering helper and are limited to the local development environment.

mdcp-feature-level