The Agent Skills Directory

[SAFE]: The skill defines a structured design process with explicit 'gates' and approval checkpoints. It requires the user to review and approve designs, specifications, and implementation plans before any downstream tools are invoked.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests untrusted data from project instructions and files.
Ingestion points: The 'Shared intake' section in SKILL.md instructs the agent to read project instructions, relevant files, and recent changes.
Boundary markers: The skill includes logic to 'Separate facts the agent can inspect from decisions the user owns' and requires explicit approval for designs and specifications.
Capability inventory: The skill can write files to the repository (project/specs/, project/research/) and invoke other implementation-related skills.
Sanitization: Risks are mitigated through mandatory human review of all generated specifications and plans before they are finalized or handed off to implementation tools. (Severity: SAFE)
[SAFE]: References to external GitHub repositories (obra/superpowers, mattpocock/skills, cursor/plugins) are used for methodology attribution and documentation purposes. No automated downloads or remote code executions are triggered by these references.

shape-design