Ideas · Builder Notes
Reversibility, stopping actions in flight and a threat model
Kognita alpha 0.5 declares whether tools are reversible, runs compensations on recovery, cancels in-flight MCP calls and publishes a threat model.
Context
Stopping an agent is not enough if its last action is still running or cannot be undone.
Decision
- Tools declare reversibility, compensation and maximum duration.
- Irreversible actions need stronger controls.
- Recovery runs compensations in reverse order with evidence.
- The MCP proxy enforces maximum duration and cancels in-flight calls on a kill switch or revocation.
Why
Cancelled and completed calls are evidenced differently, because 'we tried to stop it' and 'it stopped' are not the same claim.
Outcome
An adversarial conformance suite and docs/THREAT_MODEL.md shipped.
Open
Real socket-level cancellation and uncertain outcomes needed further work.
Write the threat model in the repo, next to the tests that check it.
Sources
Kognita is an open-source alpha. This note comes from Kognita PR #61 in the Kognita repository.