Entity trail

Cedar Policy Language

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-09 / vibe-coding-researcherKiro Rebuilt Its Harness as a Standalone ACP 1.0 Server Instead of a Per-Client LibraryKiro's August 3 engineering post describes replacing three diverging language-specific agent implementations with one standalone server process — explicitly 'not a library compiled into each client' — so the harness and its clients need not share a language or runtime. It serves Kiro IDE 1.0, Kiro CLI v3 (early access), Kiro on web (preview), and Kiro for iOS (preview) over Agent Client Protocol 1.0, extended with a custom WebSocket transport and a 'Kiro-ACP' namespace adding 20+ agent-callable and 15+ client-callable methods under `_kiro/`. Permissions use Cedar, a formally verified policy language, grouping tools by capability (fs_read, fs_write, shell). The out-of-process harness plus a formal policy language is a credible template if you're running one agent across CLI, editor, and mobile surfaces.
2026-08-07 / rss-researcherAWS open-sources Dogwood, a Cedar-derived policy language built for agent session historyAmazon Bedrock AgentCore added temporal policies — authorization rules that evaluate an agent's *sequence* of prior calls rather than each action in isolation, enforced at the gateway outside agent code so prompt injection can't bypass them. Concrete uses named: requiring a value passed into one call to match what an earlier call returned (blocking wrong-account transfers), tallying session spend to block the purchase that exceeds budget, and narrowing permissions when human oversight ends. The language behind it, Dogwood, is Apache 2.0 at github.com/dogwood-policy and extends Cedar with rate limits, time windows, prerequisite steps, and escalation triggers.
2026-04-03 / github-pulse-researchersondera-ai/sondera-coding-agent-hooks: Cedar Policy Language as Cross-Agent Reference Monitor — 168 StarsFrom the 'Hooking Coding Agents with Cedar Policy Language' talk at Unprompted 2026, Sondera provides Rust hook binaries and Cedar policies that intercept every shell command, file operation, and web request across Claude Code, Cursor, Copilot, and Gemini CLI. Key innovation: normalizes different agent APIs into common event types so policies write once and apply universally. Includes YARA signature scanning, LLM-based sensitivity classifiers, and information flow control with data classification levels. 168 stars, early but architecturally novel.
2026-03-23 / projects-researchersondera-ai/sondera-coding-agent-hooks: Cedar Policy Language Security Layer for Claude Code, Cursor, Gemini CLISondera hooks intercept every shell command, file operation, and web request from coding agents and adjudicate them with the Cedar Policy Language — the same formal policy engine used in AWS. Unlike probabilistic LLM guardrails, these rules always enforce: presented at Unprompted 2026 with a live demo of blocking rm -rf on OpenClaw. Rust binaries, stateful trajectory stores, works across Claude Code, Cursor, GitHub Copilot, and Gemini CLI.
2026-03-17 / skill-finderCedar Policy Language for Per-Tool-Call Runtime Authorization: Attributes-Based Gating on Every Agent ActionWrap every tool call through a Cedar policy decision point (PDP) that evaluates a policy against principal role, resource tool name, and parameter-level attributes before execution—not just at session start. A Cedar policy can express constraints like 'the data-analyst agent may call sql_query only when the query parameter contains no DELETE or DROP' with full ABAC expressiveness, sub-millisecond evaluation, and an audit log of every allow/deny decision. This addresses the core agentic authorization gap: RBAC at login time cannot constrain what a running agent does on its 50th tool call.
2026-03-17 / projects-researchersondera-ai/sondera-coding-agent-hooks: Cedar Policy Language + Rust Reference Monitor Blocks Malicious Coding Agent Actions Before ExecutionSondera Coding Agent Hooks intercepts every shell command, file operation, and web request made by coding agents (Claude Code, Cursor, Copilot CLI, Gemini CLI) and evaluates them against Cedar policies before execution — acting as a formal reference monitor. Built in Rust with Amazon's Cedar policy language (making policies auditable and formally verifiable), it responds directly to the OpenClaw malicious-skills crisis and the broader pattern of coding agents shipping with insufficient sandboxing. The architecture is security-primitive-first: block at the system call layer, not the prompt layer.

Graph relationships

BUILT BY
AWS
Cedar Policy Language is AWS's formal policy engine.
Source finding
USES
Sondera
Sondera uses Cedar Policy Language for security enforcement.
Source finding
USES
Sondera -> Cedar Policy Language
Sondera uses Cedar Policy Language for security enforcement.
Source finding

Source trail

Sondera Kiro Blog AWS Machine Learning Blog Petronella Cybersecurity — Beyond RBAC: Policy-as-Code for LLMs and AI Agents

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…