Skip to content

Demos Completeness

LoopGlitch26 edited this page Oct 20, 2026 · 2 revisions

Completeness and tamper evidence

An agent's two calls are allowed; only the first reports a result. The second never does, as when a tool hangs, the process dies or a post-call hook doesn't fire. When the run ends the signer, which saw the decision, writes a signed capture.gap{decided_not_completed} for it. Then a copy of the bundle with one byte changed (the gap's tool_use_id, call-2 to call-1) is verified as well.

python examples/v2/demos/completeness.py   # on the dev signer
tracekit verify run.tkb --trust trust.json

What the record shows. Two policy.decision records (allow), one tool.result (call-1) and the gap naming call-2. verify.txt: Integrity: VERIFIED with a [WARN] gaps line. In the transcript the one-byte copy fails: [FAIL] manifest, [FAIL] signatures (hash does not match the event), Integrity: FAILED.

Why it matters. A trace that holds only what arrived can't tell "never ran" from "lost". The signer names the call that is unaccounted for, and nobody can move that gap to another call, or remove it, and still verify.

the example · reading a verify report

Home

Get started

How it works

Integrations

Policy and approvals

Deploy and operate

Verify the evidence

Security and privacy

Compliance

Evaluation

Demos

Help

Clone this wiki locally