Authenticate agents and require explicit conceptual lookups

[Aymericr]

Automatic hooks now stay local and remind the agent to prepare a conceptual coding problem. Explicit lookups use an authenticated, bounded request; prompts, source, manifests, paths and transcripts are excluded. The skill explains the checks and their limits to agents and humans.

Credentials are bound to the service origin, local state rejects unsafe permissions and symlinks, and authenticated requests reject redirects. Returned cards are bounded untrusted data, tied to the recorded offer, and never executed by the client. Contributions require an explicit minimized payload and separate authorization for public sharing.

Version 0.3.0 preserves installation identity during setup and adds strict card, contribution and timing receipt handling. Existing installations should update their client and skill together.

Validation: bun run check passed, including 28 adversarial client tests, schema validation, typechecking and matching plugin templates. Tests cover local-only hooks, secret canaries before network access, redirects, symlinks, origin binding, malicious and oversized responses, actual card/status response contracts, retry identity and contribution erasure. The final documentation-only correction passed bun run build.

[pascal]

I hit an error while handling your request (Model unavailable on AI Gateway free tier: Free tier users do not have access to this model. Upgrade to paid credits at https://vercel.com/d?to=%2F%5Bteam%5D%2F%7E%2Fai%3Fmodal%3Dtop-up for unrestricted…).

Please try again, rephrase, or reach out if it keeps failing.

Error id: 1f6c018f-64dc-4b85-bbb4-90f5859324c9