|
Aegis — Runtime capability enforcement for AI agents
Aegis is a deterministic capability-enforcement layer for AI agents on MCP. No LLM in the enforcement path. Every tool call evaluated against a spec, with a...
|
83 % |
|
Rebase — bug reports your coding agent can fix
The context layer between whoever finds a bug and whatever fixes it — one script tag, a causal timeline, a suspected file and line, handed to your agent.
|
82 % |
|
RuleOak — Verifiable software and resilient engineering
RuleOak builds verifiable, resilient, local-first software and open-source systems with explicit authority and evidence boundaries.
|
82 % |
|
Agentplane | Git-native control plane for coding agents · Agentplane
Agentplane puts coding agents on an approved, verifiable Git workflow with bounded authority, observed proof, recovery, and deterministic closure.
|
82 % |
|
AgentCI Guard — Review the authority behind your agent
AgentCI Guard is an experimental static linter for risky AI-agent workflows in GitHub Actions. Review token permissions, event paths and analysis limits.
|
82 % |
|
DAG-TOML | Agent Assurance Specification
DAG-TOML is a public draft specification for declarative policy, workflow, and review evidence in agent-driven development.
|
82 % |
|
Bounded Systems — a checkpoint in front of your coding agent
Your coding agent runs with your access. Bounded Systems puts one checkpoint in front of every privileged action: it checks who is asking and what they're al...
|
82 % |
|
Rebuno - Durable execution and guardrails for AI agents
Open-source durable execution, guardrails, and human-in-the-loop for production AI agents. Every LLM and tool call is recorded before it runs and checked aga...
|
82 % |
|
Roleplay | Social-engineering tests for AI agents
Find AI agent social-engineering exploits locally, fix them as a team, and prevent regressions in CI.
|
82 % |
|
AgentLineage | Runtime data lineage for AI agents
AgentLineage tracks sensitive data through agent code and blocks unsafe egress at runtime.
|
82 % |
|
agentwatchdog — host-side runtime audit for AI coding agents
Watches Claude Code, Codex CLI, aider and others from outside the agent — from /proc, where cooperation isn’t a variable. No hooks, no wrappers, no opt-out....
|
82 % |
|
WeaveFlow | Build inspectable agent workflows in Go
WeaveFlow is a graph runtime written in Go for building, executing, and inspecting LLM agent workflows.
|
82 % |
|
Trimtab — from idea to operating company
Sixteen AI specialists research your market, write the strategy, find the funding and run the outreach — and stop at every decision that is yours to make.
|
82 % |
|
Kerniv | AI-Native Hardware Validation & Engineering Intelligence
Kerniv is building an AI engineering platform for electronic hardware validation, automated testing, firmware debugging, and physical-system diagnostics.
|
82 % |
|
Reflow — Your agents test your code, not your product
Give your coding agent the product context and runtime evidence a code diff misses. Run real journeys, compare snapshots and review what changed before you m...
|
81 % |
|
Riddle - Hosted Execution for AI Agents
Hosted execution infrastructure for AI agents: browsers, server previews, screenshots, console logs, network HAR, and proof receipts.
|
81 % |
|
Stord | Undo agent file changes
Open-source CLI that records file reads and writes from coding agents, attributes them to the agent, and lets you reverse changes from the terminal.
|
81 % |
|
Sverklo - Run Repo-Memory Proof Before Your Agent Edits
Run a proof receipt on a real repo before your coding agent edits. Sverklo runs locally by default, keeps telemetry off by default, and documents the boundar...
|
81 % |
|
LoopCode — Repository control for long-running agent work
LoopCode keeps a repository's goals, work graph, decisions, and verification evidence available to every coding-agent session.
|
81 % |
|
ContextLock — integrity & provenance for AI instruction files
ContextLock gives AI agent instruction files (SKILL.md, CLAUDE.md, rules, prompt packs) the same authenticity and integrity guarantees the software world exp...
|
81 % |
Detected social media networks