The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to process external data, such as community-contributed code changes and discussion threads, which represents an attack surface for indirect prompt injection.
Ingestion points: Community-submitted code-change.diff files and threads in the app's Discussion section (docs/launch-gate.md).
Boundary markers: The platform uses automated preflight checks, deployment reports, and safety gate findings to validate configurations (docs/preflight.md).
Capability inventory: The agent can invoke platform services (ctx.services.invoke), manage treasury assets, and publish new code releases (docs/capability-ladder.md, docs/lifecycle.md).
Sanitization: Mitigations include a human-in-the-loop approval process by owners or token holders, and automated 'Gauntlet' audits that enforce script-src constraints and reproducible builds (docs/launch-gate.md).
[COMMAND_EXECUTION]: The instructions involve the use of the vendor's bounded command-line utility for managing app lifecycles and deployments.
Evidence: Execution of commands such as bounded site deploy, bounded oapp preflight, and bounded services search (docs/preflight.md, docs/lifecycle.md).
[EXTERNAL_DOWNLOADS]: The skill facilitates interaction with external APIs via a managed payment relay.
Evidence: The agent can utilize the X402_FETCH service to interact with third-party APIs, which is controlled by locked egress boundaries and vendor-managed infrastructure (docs/capability-ladder.md).

openapps