The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's research phase involves ingesting codebase content which could contain malicious instructions aimed at influencing the agent's implementation steps.
Ingestion points: The agent uses view_file, grep_search, and find_by_name to read workspace content in Phase 2.
Boundary markers: No explicit delimiters or instructions are provided to the agent to treat codebase content as untrusted data.
Capability inventory: The agent has access to run_command, write_to_file, and replace_file_content to execute its generated plans.
Sanitization: There is no evidence of sanitization or filtering logic for the data retrieved from the workspace.
[COMMAND_EXECUTION]: The methodology explicitly directs the agent to define and execute shell commands for verification (e.g., test suites, linters, and compilers) using the run_command tool during the Plan and Implement phases.
[DYNAMIC_EXECUTION]: The skill includes a shell hook script (hooks/prompt-hook.sh) that dynamically generates system directives to be injected into the agent's prompt context based on the current state of session artifacts in the workspace.

qrspi-methodology