ur-agent 1.58.1 → 1.59.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md
CHANGED
|
@@ -1,5 +1,31 @@
|
|
|
1
1
|
# Changelog
|
|
2
2
|
|
|
3
|
+
## 1.59.0
|
|
4
|
+
|
|
5
|
+
- Added `ur sources`, a claim-to-source ledger. `wrapUntrusted` already stamped
|
|
6
|
+
every untrusted block with a nonce and a source label, but discarded both the
|
|
7
|
+
moment the block reached the model, so there was no way to audit what web or
|
|
8
|
+
MCP content an answer was built on. Recording happens inside `wrapUntrusted`
|
|
9
|
+
itself — the single choke point every untrusted block passes through, so the
|
|
10
|
+
ledger cannot miss one. `--check "<span>"` reports which fetched source
|
|
11
|
+
contains a span, and says plainly when none does, which is the useful signal:
|
|
12
|
+
that claim was not grounded in anything UR retrieved. This is the automatic
|
|
13
|
+
counterpart to the existing `/claim-ledger`, which records claims a human or
|
|
14
|
+
the agent asserts by hand; `ur sources` records what actually entered context
|
|
15
|
+
without anyone having to remember to log it.
|
|
16
|
+
- The ledger is in-memory and capped. Persisting it would create a second
|
|
17
|
+
on-disk store of third-party content — including whatever a prompt-injection
|
|
18
|
+
attempt put there — with its own retention and deletion obligations.
|
|
19
|
+
- Added `ur grade-trajectory`, which grades a run on how it worked rather than
|
|
20
|
+
what it concluded: unverified changes, edits to files never read, destructive
|
|
21
|
+
commands, and loops on identical failures. Every rule is deterministic and
|
|
22
|
+
read from the transcript; no model grades another model, because a judge that
|
|
23
|
+
can hallucinate turns a CI gate into a coin flip.
|
|
24
|
+
- `--min-score` sets `process.exitCode`, so the gate genuinely fails a CI step.
|
|
25
|
+
Returning an `exitCode` field is silently ignored by `runLocalTextCommand`,
|
|
26
|
+
which exits with `process.exitCode ?? 0` — the first implementation printed
|
|
27
|
+
FAILED and exited 0.
|
|
28
|
+
|
|
3
29
|
## 1.58.1
|
|
4
30
|
|
|
5
31
|
- Unified vision-capability detection behind
|