The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to research and ingest data from untrusted or external sources such as the repository codebase, product specifications, and external issue trackers (GitHub, Linear).
Ingestion points: The agent reads PRODUCT.md, codebase source files (e.g., app/src/workspace/mod.rs), and descriptions from Linear tickets or GitHub issues.
Boundary markers: Absent. There are no explicit instructions or delimiters provided to ensure the agent ignores or sanitizes instructions that might be embedded within the files being researched.
Capability inventory: The agent is authorized to write new files to the specs/ directory and execute developer CLI tools (git, gh).
Sanitization: Absent. The skill does not mention sanitizing or escaping content retrieved from external files or issue descriptions before using it to generate implementation plans.
[COMMAND_EXECUTION]: The skill provides instructions for the agent to use standard developer command-line tools to gather repository metadata and interact with project management services.
Evidence: The instructions explicitly mention using git rev-parse HEAD to capture commit SHAs and the gh CLI or Linear MCP tools for managing issues.
Guardrails: The skill contains a specific instruction to only create new tickets or issues when the user "explicitly asks for one," providing a manual approval step for network-connected operations.

write-tech-spec