The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection via the project design document.
Ingestion points: Processes specifications from the docs/plans/<feature>/design.md file.
Boundary markers: Absent; the runner prompt does not utilize explicit delimiters or warnings to separate user-provided data from agent instructions.
Capability inventory: The skill described in SKILL.md has capabilities to write algorithmic code via mcp__speed-run__generate_and_write_files and execute shell commands for testing and environment setup.
Sanitization: Absent; content from the design document is interpolated into implementation plans and prompts without validation or escaping.
[DATA_EXFILTRATION]: The skill transmits project technical data to an external service for processing.
Design specifications and contract prompts are sent to Cerebras, a third-party hosted LLM provider, to facilitate code generation.
[COMMAND_EXECUTION]: The skill performs various system and version control operations.
Uses git worktree to create isolated environments for parallel execution and git branch for version management.
Executes diff to compare implementations across runners and runs local test suites to verify generated code.
Utilizes vendor-specific resources fresh-eyes-review:skills and scenario-testing:skills for quality assurance and verification.

showdown