@metamynd/agentsafe-guard 0.6.4 → 0.6.6
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/README.md +14 -1
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -25,7 +25,7 @@ re-implement the gate:
|
|
|
25
25
|
([markdown](https://metamynd.ai/specs/magp-v1.0.md))
|
|
26
26
|
|
|
27
27
|
It defines agent identity, the canonical signed message (§8.3), the sixteen-stage order of
|
|
28
|
-
checks (§8.5), all
|
|
28
|
+
checks (§8.5), all 64 reason codes (Appendix A), delegation narrowing (§5.4), and evidence
|
|
29
29
|
you can verify offline without MetaMynd (§13.4). If you are writing a client in a language
|
|
30
30
|
other than JavaScript, read §8.3.3–8.3.5 first: key encoding, number stringification and
|
|
31
31
|
signed-vs-sent field identity each surface only as `SIGNATURE_INVALID`.
|
|
@@ -99,6 +99,19 @@ so the identical "mismatch → not satisfied" rule let a prohibition like `payAm
|
|
|
99
99
|
unit USD` be silently skipped by declaring any other currency, including a mere case
|
|
100
100
|
difference (`'usd'` vs `'USD'`). The currency comparison is also now case-insensitive.
|
|
101
101
|
|
|
102
|
+
**0.6.5 — `IDENTITY_KEY_MISMATCH`, a new reason code (now 63 total).** The backend gate
|
|
103
|
+
now cross-checks a resolved agent's registered public key against the key embedded in its
|
|
104
|
+
own self-certifying DID (`did:hedera` / `did:key`) before trusting a signature against it —
|
|
105
|
+
free, local, no network round-trip, since the DID already carries its own proof. Nothing in
|
|
106
|
+
this package's own evaluation changes; documented here because the count in this README and
|
|
107
|
+
the linked spec moved.
|
|
108
|
+
|
|
109
|
+
**0.6.6 — `SPEND_PATTERN_ANOMALY`, a new reason code (now 64 total).** The backend gate can
|
|
110
|
+
now escalate an otherwise-permitted decision whose amount deviates sharply from an agent's
|
|
111
|
+
OWN recorded spend history — a baseline the platform's trust-graph engine learns, not a
|
|
112
|
+
threshold a human pre-sets. Opt-in (`SPEND_ANOMALY_MODE=on`), off by default. Nothing in this
|
|
113
|
+
package's own evaluation changes; documented here because the count moved again.
|
|
114
|
+
|
|
102
115
|
```yaml
|
|
103
116
|
# .github/workflows/governance.yml
|
|
104
117
|
name: Governance
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@metamynd/agentsafe-guard",
|
|
3
|
-
"version": "0.6.
|
|
3
|
+
"version": "0.6.6",
|
|
4
4
|
"description": "Zero-dependency runtime governance for any Node AI agent \u2014 gate tool calls through MetaMynd/AgentSafe (allow / block / escalate) against the agent's mandate, enforced Standards, and SOPs. Ed25519-signed, deterministic, fail-closed.",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"main": "./agentsafe-guard.mjs",
|