The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes data from external Locus and Stripe endpoints, which presents a surface for indirect prompt injection.
Ingestion points: The agent reads account.registration.authorizationUrl from API responses and processes user-facing messages from Stripe humanHandoff. It also fetches search results from the capability catalog.
Boundary markers: The instructions explicitly command the agent to treat fetched content as untrusted data and ignore any embedded instructions that redirect credentials or scope.
Capability inventory: The skill is capable of performing network requests to vendor APIs, writing secrets to protected local files (~/.config/locus/), and executing local shell commands.
Sanitization: The skill emphasizes strict adherence to the documented protocol and the use of cryptographic RNGs to prevent manipulation by external inputs.
[COMMAND_EXECUTION]: The skill utilizes local shell commands to generate secure registration tokens.
Evidence: The instructions direct the agent to run python3 -c 'import secrets; print(secrets.token_urlsafe(24))' to generate a transient registration token.
[EXTERNAL_DOWNLOADS]: The skill references and fetches content from vendor-controlled domains.
Evidence: Fetches authorization links and continuation URLs from api.paywithlocus.com and auth.agentid.com during the signup and OAuth flows. References to external documentation at paywithlocus.com are mirrored in the local repository for verification.

locus-setup