The Agent Skills Directory

[SAFE]: The skill is composed entirely of Markdown instructions and reference documents. It contains no scripts, executables, or code blocks that could be run on the host system.
[SAFE]: No network operations (e.g., curl, wget, fetch) or remote resource downloads are present in the instructions or reference files.
[SAFE]: There are no hardcoded credentials, API keys, or attempts to access sensitive system paths (e.g., .ssh, .aws, .env).
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted user input and project context (files, tickets, prior conversations) to generate a specification.
Ingestion points: Phase 1 (Intake) instructions direct the agent to read user-provided ideas and external context files.
Boundary markers: The skill does not explicitly instruct the agent to use XML delimiters or specific 'ignore' tags for untrusted content, but it does enforce a multi-phase human-in-the-loop review process.
Capability inventory: The skill has no capability to execute code, perform network requests, or modify the file system beyond writing the resulting specification text.
Sanitization: No explicit sanitization of user input is mentioned, but the skill's primary rule ('No code until approved') acts as a significant safety barrier.
[SAFE]: No obfuscation techniques, hidden URLs, or dynamic context injection patterns (e.g., exclamation-mark-backtick syntax) were detected in any of the skill files.

brainstorming