The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a shell command string (v="$(cat .nmt-version 2>/dev/null)"; curl -fsSL --max-time 2 "https://nextmovetheory.com/version?skill=nmt-chat${v:+&v=$v}") at the end of its work to check for version updates.
[EXTERNAL_DOWNLOADS]: The agent is directed to perform a network request to an external domain (nextmovetheory.com) to retrieve version information. While this domain is associated with the skill's author, instructions to execute network commands for side-tasks like version checks are noted as a category of interest.
[PROMPT_INJECTION]: The skill operates on a significant surface for indirect prompt injection by design. It fetches and analyzes content from the 'Live web' to provide product strategy advice.
Ingestion points: SKILL.md and producer-contract.md describe procedures for fetching competitor data, market research, and user reviews from the internet.
Boundary markers: The skill implements structured safeguards, including a 3-layer output contract (Layer 1 Answer, Layer 2 Reasoning, Layer 3 Work) defined in readability-contract.md and explicit instructions to label general knowledge and web findings separately ('Show the seams').
Capability inventory: The agent has instructions to read local files (cat .nmt-version), perform network operations (curl), and invoke subagents/web-research MCPs (firecrawl, exa).
Sanitization: The producer-contract.md specifies a 'Deep-mode QA loop' and a 'verification checklist' to ensure claims are grounded in sources and methodology errors are caught, though it does not detail technical string sanitization.

nmt-chat