The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from project files such as ROADMAP.final.md, EXECUTION.md, and ACCEPTANCE.md to update canonical documentation.
Ingestion points: Files located in docs/<feature>/, specifically ROADMAP.final.md, PLAN.final.md, and EXECUTION.md.
Boundary markers: None explicitly defined to separate data from instructions during the triage and archival process.
Capability inventory: File reading, writing, and deletion (git rm); execution of git commands and arbitrary project generators/tests.
Sanitization: No sanitization or validation of input file content is mentioned before interpolation into documentation or the new roadmap.
[COMMAND_EXECUTION]: The skill instructs the agent to run project-specific tools to update documentation.
Evidence: Instructions include "run the generator and commit what it produces" and "Run any tests that enforce generated docs".
Risk: If the project repository contains malicious scripts masquerading as generators or tests, the agent will execute them with the user's permissions.
[DYNAMIC_EXECUTION]: The skill relies on runtime execution of project-defined commands to maintain documentation integrity.
Evidence: The agent is tasked with running unknown "generators" and "tests" that vary per project environment.

close-roadmap