The Agent Skills Directory

[DATA_EXFILTRATION]: The skill documents the --deliver webhook:<url> feature, which allows the output of any command (potentially containing private repository data or local file metadata) to be POSTed to an arbitrary external URL.
[DATA_EXFILTRATION]: The feedback command provides a mechanism to transmit text data to a remote endpoint configured via the GITHUB_CONTENTS_FEEDBACK_ENDPOINT environment variable.
[COMMAND_EXECUTION]: The 'Automatic learning' loop (recall, teach, playbook, amend) records command history and synthesizes 'playbooks'. These playbooks are sequences of commands that the agent is instructed to replay, which constitutes dynamic execution of stored instructions.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its learning loop. Specifically, the teach and playbook commands ingest natural language queries and choreography that influence future agent behavior.
Ingestion points: Data enters the system via the teach command's --query and --playbook-json arguments, and is stored in a local data.db and teach.log.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are present in the documentation for processed data.
Capability inventory: The skill can execute shell commands, perform network requests (GitHub API, webhooks), and write to the local file system (fetch, sync-dir, --deliver file:, and state/config directories).
Sanitization: There is no evidence of sanitization or validation of the stored 'learnings' before they are used to influence agent decisions via the recall command.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of external binaries (github-contents-pp-cli and github-contents-pp-mcp) using npx and go install from the author's (mvanhorn) namespace and GitHub repositories.

pp-github-contents