Examples

Worked examples live in the repository-level examples folder.

Coding agents should ingest examples/connector-agent-contract.json before writing a connector. It lists the required source-context fields, observed-evidence fields, allowed agent steps, and non-negotiable boundaries.

ABAG Connector Context

The ABAG example shows a source-only Abstract Bayesilisk Action Graph. It maps concrete connector actions such as create-invite, revoke-invite, and accept-invite-route onto universal typed tokens such as identifier.invitation_token and state.revoked. Bayesilisk can compose a reusable revoked-token replay motif from that graph while the connector remains responsible for executing concrete app fixtures and returning observed evidence.

Cal.com Connector Evidence

The Cal.com example shows Bayesilisk used with its general core and an app-specific connector. The connector follows the documented contract: it supplies source facts, explicit proposal rules, connector actions, expected statuses, and observed local evidence. Bayesilisk expands only those supplied rules, then verifies expected-versus-observed facts.

Captured evidence is included for Cal.com commit 180ede28f0bddf2738933a6e60a8e80f6116d7da from https://github.com/calcom/cal.com.

The checked-in artifacts show:

  • route source facts with explicit proposal rules;

  • sequence source context with a connector-declared action graph;

  • 7 Bayesilisk-generated proposals;

  • 7 local connector observations;

  • 7 connector-observed findings: 6 unknown/stale rescheduleUid route mutations where expected 404 was observed as 200, plus 1 cancelled booking replay workflow where expected 409 was observed as 200;

  • app-only JSON, Markdown, and issue-payload reports;

  • upstream issue and fix-PR references in examples/calcom/upstream-outcomes.md.

The upstream outcome references matter: an open fix PR linked to a Bayesilisk finding and then approved by a human reviewer is stronger validation than a closed issue by itself, because it shows one human contributor found the report concrete enough to implement a targeted change and another human reviewer found the fix correct enough to approve.

Explanatory prose in connector context may help Grassmann attention route and rank the investigation, but proposal expansion comes from explicit rules and verdicts come from observed evidence.