https://github.com/rybkr/bridge-keeper
Security constrained agent runtime environment
https://github.com/rybkr/bridge-keeper
Last synced: 3 months ago
JSON representation
Security constrained agent runtime environment
- Host: GitHub
- URL: https://github.com/rybkr/bridge-keeper
- Owner: rybkr
- Created: 2026-02-02T22:26:49.000Z (5 months ago)
- Default Branch: main
- Last Pushed: 2026-03-25T15:54:20.000Z (3 months ago)
- Last Synced: 2026-03-26T17:41:16.265Z (3 months ago)
- Language: Go
- Homepage:
- Size: 131 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 2
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Bridgekeeper
A security-constrained agent runtime that enforces capability-based security for tool use.
Bridgekeeper treats the agent as an untrusted program: it can propose tool calls, but the runtime mediates execution through a policy engine, local sandbox validation, human approval for `ask` decisions, and structured audit logging. Tool inputs are normalized before execution, and sensitive tool output is redacted before it is handed back to the model.
Current state:
- Policy evaluation for tool/action/capability matching is implemented.
- Local sandbox enforcement currently focuses on workspace-bounded filesystem access, argument validation, and output-size limits.
- Audit logging is structured JSONL.
- Sensitive output redaction and simple taint detection are implemented.
- Network sandboxing and deeper information-flow controls are still incomplete.
## Project Structure
```
bridgekeeper/
├── cmd/bridgekeeper/ # CLI entrypoint
├── internal/
│ ├── agent/ # Provider-specific agent adapters (Gemini, etc.)
│ ├── runtime/ # Core mediation and runtime loop
│ ├── policy/ # Policy engine and YAML loader
│ ├── tools/ # Typed tool implementations grouped by capability
│ ├── sandbox/ # Workspace and payload validation below policy
│ ├── redact/ # Secret redaction and sensitivity classification
│ ├── audit/ # Structured audit trail logging
│ └── hitl/ # Human-in-the-loop approval
├── policies/ # YAML policy files
├── testdata/
│ ├── adversarial/ # Prompt injection and adversarial fixtures
│ └── workflows/ # End-to-end workflow scripts
└── docs/ # Design doc and evaluation report
```