The Agent Skills Directory

[COMMAND_EXECUTION]: The skill runs shell commands for environment discovery (ls), validation execution (node scripts/coherence.mjs), and state persistence (git pull, git push).
[EXTERNAL_DOWNLOADS]: The skill uses npx to execute tools like js-yaml and tsx, which involves fetching packages from the NPM registry if they are not available locally.
[DYNAMIC_EXECUTION]: The skill executes local scripts and uses npx tsx to evaluate inline code for schema validation.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project artifacts which are passed to LLM agents, representing a vulnerability to indirect prompt injection.
Ingestion points: Files identified at product-context.yaml, product/index.yaml, and artifacts within openspec/changes/ or .dev-workflow/.
Boundary markers: The skill does not explicitly define delimiters for the ingested content in its instructions to the agents.
Capability inventory: The skill can write to the filesystem, push to remote git repositories, and execute shell commands via node or npx.
Sanitization: The skill performs syntactic validation using js-yaml and mechanical coherence checks to ensure data integrity before agent processing.

aep-validate