A Git-native self-evolving skill bay for agents: decentralised skill discovery, collaboration, and sync with signed provenance, PoW-gated publishing, policy-controlled execution, and compartmentalised trust.
This is a V0 prototype implementing:
- CLI for skill creation, publishing, discovery, and execution
- Git integration for skill versioning
- Cryptographic signing and PoW for anti-spam
- Local and central policy evaluation (central checks deferred for execution)
- Basic in-memory discovery (P2P stubbed)
- Simple subprocess execution (no sandboxing)
- Web UI stub (basic server, no advanced features)
- Reputation system with reviews and credibility scoring
Not implemented in V0:
- Full P2P network with DHT
- Advanced consensus mechanisms
- Sandboxed execution
- Semantic deduplication
- Enterprise authentication
- Advanced economics/incentives
cargo build --release./target/release/harbour --helpinit <path>: Initialize workspacecreate-skill <name> <description> <path>: Create a new skillpublish <path>: Publish a skill with Git commit, sign, PoWdiscover <query>: Find skills (in-memory)verify <skill_id>: Check signature and PoWinstall <skill_id> <path>: Clone and install skill if policy allowsexecute <skill_id> [args]: Run skill if policy allowsshow-policy-decision <skill_id>: Check execution policyset-policy <blocked> <min_pow>: Configure local policyset-central-policy <banned>: Configure central policy
patch <path> <description>: Create a patch (Git-based, no PR)pin <skill_id>: Pin a skill for availability (stub)web: Start basic web UI for skill discoveryreview <skill_id> <rating> <comment>: Review a skillp2p: Start P2P node (stubbed network behavior in V0)
After building:
# Run tests
cargo test
# Create a test skill
mkdir -p test_skill
./target/release/harbour create-skill "test" "A test skill" test_skill/
# Publish it
./target/release/harbour publish test_skill/
# Discover it
./target/release/harbour discover "test"
# Verify it
./target/release/harbour verify <skill_id>
# Install it
./target/release/harbour install <skill_id> install_path/
# Execute it (requires skill to have executable entry_point)
./target/release/harbour execute <skill_id> arg1 arg2See ARCHITECTURE.md for details.
See SECURITY.md for trust model and current limitations.