The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches the blacklane-pp-cli tool and supporting libraries from the author's GitHub repository (github.com/mvanhorn) and the NPM registry (@mvanhorn/printing-press-library).
[COMMAND_EXECUTION]: Instructions direct the agent to install and execute a binary CLI tool (blacklane-pp-cli) using npx and go install commands.
[DATA_EXFILTRATION]: The CLI provides a --deliver webhook:<url> flag which allows the output of any command (potentially containing sensitive pricing history or trip logs) to be sent to an arbitrary external URL.
[DATA_EXFILTRATION]: The skill includes a feedback command that can send local usage data and session notes to a remote endpoint if the BLACKLANE_FEEDBACK_ENDPOINT environment variable is configured.
[INDIRECT_PROMPT_INJECTION]: The skill implements a persistent "learning loop" using recall, teach, and playbook amend commands that store session data and instructions into a local SQLite database to influence future behavior. This creates a surface where malicious user input could be "taught" to the agent and replayed in subsequent sessions.
Ingestion points: User queries and instructions processed by the recall, teach, and playbook amend commands.
Boundary markers: The instructions lack programmatic boundary enforcement for the stored knowledge, instead relying on manual agent guidelines.
Capability inventory: The skill has access to shell execution via the Read Bash tool and network exfiltration via the webhook and feedback CLI features.
Sanitization: The skill recommends manual stripping of PII by the agent but does not implement automated sanitization or strict schema validation for learned data.

pp-blacklane