The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external data (roadmap and plan files) to generate a self-contained /goal prompt. This creates a surface for indirect prompt injection, where malicious instructions embedded in these files could be propagated to and executed by the downstream 'executor' agent.
Ingestion points: The skill reads roadmap .mdx files and additional markdown plan files provided via the --plan argument in SKILL.md.
Boundary markers: The skill relies on a structured skeleton (Objective, Scope, Verify gates, etc.) to organize the output, but it does not include explicit delimiters or 'ignore embedded instructions' warnings to protect the downstream agent from malicious content in the source files.
Capability inventory: The skill instructions involve reading local filesystem content and executing shell commands (wc -c).
Sanitization: No explicit sanitization, filtering, or validation of the input content is performed before it is distilled into the final prompt.
[COMMAND_EXECUTION]: The process instructions in SKILL.md require the agent to write the generated prompt to a temporary file and execute the wc -c shell command to enforce a hard 4000-character limit. While this is a standard utility used for a benign purpose, it involves direct interaction with the system shell.

jackin-goal-prompt