The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides templates for defining functional verification commands (e.g., curl calls and CLI invocations) within a truths field. These commands are intended to be executed by a verification tool (loom check) to confirm functional integration. The instructions include explicit security constraints, such as the NO PATH TRAVERSAL rule which strictly prohibits the use of ../ in file paths.
[PROMPT_INJECTION]: The skill defines a surface for processing integration requirements into executable verification logic, which is susceptible to indirect prompt injection if untrusted source data is used.
Ingestion points: Loom plan stage metadata and feature requirements provided in the agent's context (SKILL.md).
Boundary markers: Uses structured YAML templates and specific refinement questions to validate logic; however, it lacks explicit delimiters for untrusted external content.
Capability inventory: Uses Bash, Write, Edit, and Grep tools to execute verification commands and inspect codebase patterns.
Sanitization: Prohibits path traversal sequences and provides strict formatting rules for YAML description fields to prevent syntax-based injection.

loom-wiring-test