Declaw
A sandbox platform combining Firecracker execution with external policy and credential mediation.
Routing Summary
Keep as a comparison candidate when egress policy, credential injection, and auditability drive provider selection. Compare Runloop Devboxes and Agent Security Model.
Verified documentation
Declaw advertises an external proxy, network filtering, credential vaulting, guardrails, and snapshots. Its security and prompt-injection defenses are vendor claims, not a demonstrated guarantee. A production evaluation must test permitted and denied actions and whether secrets can leak through responses or logs. Source: official documentation, checked 2026-09-25
Evidence status
Documentation review only; no account setup, workload benchmark, or isolation test was performed. Keep rules, skills, approvals, source history, and proof in Kevin's control plane. Provider execution must remain replaceable. See Agent Sandboxes for the comparison contract.
Package snapshot: see the dated registry version in Agent Sandboxes.
Timeline
- 2026-09-25 | Recovered this provider identity from the July 7 test2 snapshot and rewrote it against current primary documentation. Source: test2 snapshot