The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes dynamically detected test runners and setup commands (such as npm test or pytest) within a Bash subagent during the verification stage (Stage 7). These commands are derived from local project configuration files like package.json or pyproject.toml via the project detection logic in references/project-detection.md.
[EXTERNAL_DOWNLOADS]: During its pre-flight dependency check, the skill identifies if the ship skill is missing and recommends installation from the author's repository (slamb2k/mad-skills) using the npx skills add command. This is a vendor-owned resource intended to extend the pipeline's capabilities.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads untrusted data from the local codebase (e.g., source code, documentation) and user-provided plans, which are then substituted into prompts for subagents. A subagent receiving a plan or code containing malicious instructions could potentially be influenced to perform unauthorized actions using its available tools.
Ingestion points: User-provided plans passed as arguments to the build command; codebase exploration results from Stage 1 (found in references/stage-prompts.md).
Boundary markers: Prompts in references/stage-prompts.md use structured headers like ## Plan and ## Exploration Findings to delimit external data.
Capability inventory: Subagents used by the skill possess Write, Edit, and Bash capabilities across various stages (Implementation, Fix, Verify).
Sanitization: Content is substituted directly into templates; no explicit sanitization or instructions to ignore embedded commands within the ingested codebase data are provided.

build