The Agent Skills Directory

[SAFE]: The skill is designed for documentation management and initiative decomposition within a local project environment. It reads from local files like seed.md and decisions.md and writes to roadmap.md. All operations are limited to the filesystem with no evidence of network access, external downloads, or execution of arbitrary code.
[DATA_EXFILTRATION]: The skill accesses local files within the docs/threads/ directory to gather context for generating roadmaps. This behavior is confined to the intended project directory structure and does not involve accessing sensitive system files, environment variables, or private user data. No network calls or exfiltration patterns were detected.
[PROMPT_INJECTION]: The instructions provide a clear, logical framework for the agent to follow, including pre-flight checks and refusal conditions for ambiguous inputs. No instructions were found that attempt to bypass safety filters or override the agent's core behavioral constraints.

roadmap