The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs the agent to install dependencies using npx -y @mvanhorn/printing-press-library and go install github.com/mvanhorn/printing-press-library/.... These resources are hosted by the vendor 'mvanhorn'.
[COMMAND_EXECUTION]: The skill operates by executing the lever-pp-cli binary to perform tasks like listing job postings, verifying installation with doctor, and managing local state.
[REMOTE_CODE_EXECUTION]: The skill implements a 'Playbook' system where the agent is instructed to 'replay Playbook.steps in order' from JSON files. These steps consist of shell commands that are dynamically executed based on content stored in a local SQLite database or provided via external playbook files.
[DATA_EXFILTRATION]: The CLI includes a --deliver webhook:<url> flag that allows the agent to POST command results to any provided URL. While a legitimate feature for integration, it provides a functional capability for sending data to non-whitelisted external domains.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes untrusted job posting data from the public Lever API. This data is incorporated into the 'learning loop' and local 'recall' store, which influences the agent's future decision-making and command execution without explicit sanitization or boundary markers between data and instructions.
Ingestion points: Public Lever API (api.lever.co) and local SQLite database (data.db).
Boundary markers: Output is delimited within JSON envelopes (meta and results fields), but there are no instructions to ignore embedded commands within the results.
Capability inventory: Execution of CLI subcommands, file writing via --deliver file:, and network POSTing via --deliver webhook:.
Sanitization: No explicit sanitization or filtering of external API content is documented.

pp-lever