The Agent Skills Directory

[PROMPT_INJECTION]: The skill directly interpolates the $ARGUMENTS variable into the prompt string for the /ralph-loop:ralph-loop tool. This allows a user to inject malicious instructions that could override the intended behavior of the implementation loop or bypass safety constraints.
[PROMPT_INJECTION]: The skill processes data from external specification files and treats their content as authoritative instructions for code implementation, creating a surface for indirect prompt injection.
Ingestion points: Specification files are read from specs/$ARGUMENTS/spec.md and specs/{spec-name}/spec.md.
Boundary markers: There are no delimiters or specific instructions (e.g., 'ignore embedded instructions') used to separate specification content from the agent's core instructions.
Capability inventory: The agent has the capability to execute shell commands (running tests), perform git operations (commit and push), and trigger deployments.
Sanitization: The skill does not validate, escape, or filter the content of the specification files before the agent acts upon them.
[COMMAND_EXECUTION]: The execution flow explicitly commands the agent to "Run all tests (unit, integration, browser, visual)". This grants the agent permission to execute arbitrary code contained within the project's test suite.
[DATA_EXFILTRATION]: The instructions to "Commit and push changes" and "Deploy if required" provide a functional pathway for project data and source code to be transmitted to external servers and repositories.

speckit-implement