PrismorSec

PrismorSec/prismor

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

296 stars 27 forks 296 watchers Python Apache License 2.0
agent-control-plane agent-security agents ai-agents ai-security claude-code control-plane cursor cybersecurity devsecops guardrails llm-security mcp owasp prompt-injection secrets-management security skill-scanner supply-chain-security tool-calls
View on GitHub Website
34 Open Issues Need Help Last updated: Aug 19, 2026

Open Issues Need Help

View All on GitHub
redact upstream HTTP 4xx/5xx error bodies to prevent secret leakage about 11 hours ago
bug good first issue area:cloaking area:proxy security
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Forward signed audit-record provenance through telemetry sinks about 15 hours ago
enhancement good first issue help wanted area:runtime security
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Decide block/warn policy for high-FP-risk operations (T1-T15 gap-hunt residuals) about 15 hours ago
help wanted question area:runtime
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
scoped-agent denials over-block: only measured cause of blocking legitimate actions (utility control 100% -> 33%) about 15 hours ago
bug help wanted area:runtime
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Codex has no file-tool events at all (everything is Bash): file_read/file_write rules are unreachable about 15 hours ago
bug enhancement help wanted area:integrations
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Pre-call budget enforcement with multi-window cost limits about 15 hours ago
enhancement help wanted area:runtime roadmap
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Mid-stream cutoff for streaming agent output about 15 hours ago
enhancement help wanted area:runtime area:proxy roadmap
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Per-identity sliding-window rate limiting about 15 hours ago
enhancement help wanted area:runtime roadmap
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Runaway agent loop detection and circuit breakers about 16 hours ago
enhancement help wanted area:runtime roadmap
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
MCP per-tool budget enforcement and tool-call circuit breakers about 16 hours ago
enhancement help wanted area:mcp roadmap
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Support OPA/Rego policies alongside YAML about 16 hours ago
enhancement help wanted area:runtime roadmap
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Zero standing privileges with DPoP-bound ephemeral tokens about 16 hours ago
enhancement help wanted area:runtime security roadmap
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Agent-to-agent governance with consent escalation broker about 16 hours ago
enhancement help wanted area:runtime roadmap
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Add SECURITY.md, CONTRIBUTING.md, and CODE_OF_CONDUCT.md about 16 hours ago
documentation good first issue
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Ship official Docker image and docker-compose about 16 hours ago
enhancement good first issue help wanted area:ci area:deployment
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Kubernetes deployment via Helm chart about 16 hours ago
enhancement help wanted area:deployment roadmap
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Standalone client SDKs (Python + TypeScript) for integration about 16 hours ago
enhancement help wanted area:integrations roadmap
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Harden OSS CI: CodeQL, Scorecard, dependabot, dependency scanning about 16 hours ago
enhancement good first issue help wanted area:ci security
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Bind a cloaked secret to the destinations it may travel to about 16 hours ago
enhancement help wanted area:cloaking security
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
feat(discover): inventory the browser WebMCP surface about 16 hours ago
enhancement help wanted area:runtime area:integrations
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
MDM Fleet Deployment (Jamf/Intune): packaging, Windows support, and fleet enrollment token about 16 hours ago
enhancement help wanted area:deployment roadmap
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Decision legibility: a developer should be able to tell why an agent was allowed, blocked, or escalated in seconds about 16 hours ago
enhancement help wanted area:runtime area:dashboard roadmap
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Prismor is blind to container execution: host-absolute hook paths, and cloaking that degrades open about 16 hours ago
enhancement help wanted area:runtime area:deployment security roadmap
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Have a clear difference between prismor enterprise and OSS when installed to avoid conflicting policies about 16 hours ago
bug documentation help wanted area:runtime area:deployment
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Classification of categories in prismor replay and dashboard about 16 hours ago
enhancement help wanted area:dashboard
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
docs: session screenshot predates the Sessions redesign about 16 hours ago
documentation good first issue area:dashboard
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
optimize test suite module-state leak guard overhead (-30% test suite runtime) about 16 hours ago
enhancement help wanted area:tests performance
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
DEMO TIME 🚀 about 16 hours ago
documentation good first issue help wanted
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
prevent verdict index shift and cache poisoning in batched window analysis about 16 hours ago
bug help wanted area:runtime security
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
proxy: deferring the session snapshot by event count hides short sessions entirely about 16 hours ago
bug good first issue help wanted area:proxy performance
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
bug(provenance): cp/mv drops UNTRUSTED tag because _grams checks command string instead of source file (#379) about 16 hours ago
bug help wanted area:runtime area:provenance security
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
bug(provenance): subshell command substitution $(curl ...) bypasses provenance fetch tracking about 16 hours ago
bug help wanted area:provenance security
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
openai-agents adapter blocks in observe mode; four adapter test files never collect in a full run about 18 hours ago
bug good first issue area:integrations
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls
Cloak: writing about the placeholder syntax is blocked (docs, commit messages, tests) about 1 month ago
bug good first issue
PrismorSec/prismor
296

Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it executes: secret leaks, prompt injection, supply chain etc in a local dashboard. Agent agnostic (Claude,codex etc.)

Python
#agent-control-plane#agent-security#agents#ai-agents#ai-security#claude-code#control-plane#cursor#cybersecurity#devsecops#guardrails#llm-security#mcp#owasp#prompt-injection#secrets-management#security#skill-scanner#supply-chain-security#tool-calls