The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill implements a "playbook" system where the agent is instructed to execute shell commands retrieved from the CLI tool's local store or remote API during the recall process. This results in the dynamic execution of commands based on processed data.- [DATA_EXFILTRATION]: The CLI tool includes a --deliver webhook:<url> parameter that enables the agent to send command outputs and results to an arbitrary external HTTP endpoint.- [DATA_EXFILTRATION]: The skill describes a feedback and learning mechanism that can be configured to automatically transmit local journals, observations, and data to a remote server via the CPSC_RECALLS_FEEDBACK_ENDPOINT environment variable.- [EXTERNAL_DOWNLOADS]: The instructions guide the agent to download and install external binaries from the vendor's GitHub repository and NPM registry using npx and go install.- [PROMPT_INJECTION]: The skill processes external product recall data and local CSV files which could contain malicious instructions designed to influence agent behavior. This risk is heightened by the agent's instructions to follow "playbook" steps and "notes" derived from this ingested data. Ingestion points: CPSC API, local inventory CSVs, local learning store. Boundary markers: Absent. Capability inventory: Bash command execution, file system access. Sanitization: Minimal PII scanning is performed during the teach phase, but no sanitization of shell commands in playbooks is documented.

pp-cpsc-recalls