The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions in references/phases/polish-implement.md explicitly direct the agent to ignore standard orchestrator constraints (specifically those in rules/orchestrator-constraints.md) to allow the agent to write implementation code directly during the 'polish' track, which is documented as an intentional violation of base rules.
[INDIRECT_PROMPT_INJECTION]: The skill processes existing source code and project files, which serves as a potential vector for instructions embedded in data.
Ingestion points: Target source files read and analyzed during the 'explore', 'brief', and 'overhaul-delegate' phases (e.g., references/phases/explore.md).
Boundary markers: No explicit delimiters or instructions are provided to the agent to treat ingested code content as untrusted or to ignore embedded instructions within it.
Capability inventory: The skill possesses the ability to execute shell commands (npm run, git, gh) and modify files across the project directory.
Sanitization: The instructions do not mention any validation, escaping, or sanitization of the content extracted from external source files before it is processed by the agent.
[COMMAND_EXECUTION]: The workflow automates several shell-based development operations, including build processes (npm run build), testing runners (npm run test:run), and version control tasks using git and the GitHub CLI (gh).
[DYNAMIC_EXECUTION]: The skill facilitates the execution of local project scripts and characterization tests via npm run commands, which involves running code derived from the repository environment.

refactor