The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from multiple untrusted external sources including error logs, user testing observations, dogfood reports, and telemetry APIs. This data is used to update the product-context.yaml and influence the agent's future tasks, creating an attack surface for indirect prompt injection. \n- Ingestion points: dogfood-*.md files, lessons-learned/*.md files, and external telemetry endpoints configured in topology.routing.telemetry_sources. \n- Boundary markers: The instructions do not define explicit delimiters or 'ignore embedded instructions' warnings for these ingestion points. \n- Capability inventory: The skill possesses the capability to perform git commit/push operations, write to core product documentation, and execute shell commands via npx and curl. \n- Sanitization: No sanitization or filtering of external content is specified before the data is processed or interpolated into the product context.\n- [DYNAMIC_EXECUTION]: The skill uses npx tsx -e to execute a Node.js script string at runtime for YAML validation. While the script logic is static, it relies on runtime execution and loading of the @agentic-engineering-patterns/api package.\n- [EXTERNAL_DOWNLOADS]: The skill utilizes npx to download and run the js-yaml and tsx packages at runtime without specified versions. It also uses curl to fetch telemetry data from external endpoints configured by the user, which may point to unverified sources.\n- [COMMAND_EXECUTION]: The skill instructs the agent to perform various shell commands for project management, including Git version control (git pull, git add, git commit, git push), YAML linting via npx, and telemetry ingestion using curl and jq.

aep-reflect