hdar-canonical 1.0.0

Verifiable state continuity and provenance infrastructure for autonomous AI agents

🛡️ HDAR Protocol

Verifiable State Continuity & Provenance Infrastructure for Autonomous AI Agents

---

⚡ The Trust Gap in Agentic Workflows

As autonomous agents transition from suggestions to active execution (modifying codebases, running unit tests, deploying containerized infrastructure), they move across distinct sandbox boundaries (Local Desktop $\rightarrow$ GitHub Actions $\rightarrow$ Isolated Docker Sandboxes $\rightarrow$ Cloud VMs).

Without HDAR, organizations cannot prove:

• Identity: Which agent or platform host produced a given state change?
• Integrity: Was the workspace tampered with during transport between hosts?
• Compliance: Did the execution occur in a secure, attested environment?

HDAR (High-Density Agent Recovery) solves this by packaging workspace state, continuation points, and cryptographic signatures into verifiable Agent Capsules that chain across host transitions.

---

✨ Features & Moats

• 🔒 Ed25519 Cryptographic Signatures: Every capsule manifest is signed by the workspace owner, preventing unauthorized state modification and spoofing.
• 📦 Content-Addressed Blocks: Deduplicated file storage (blocks/<sha256[:2]>/<sha256>) ensuring fast, bandwidth-efficient state transfers.
• 🔗 Cryptographic Lineage Chains: Successor epochs reference parent manifest hashes forming a verifiably immutable Merkle chain ($E_1 \rightarrow E_2 \rightarrow E_3$).
• 📋 Clipboard Teleportation: Base64 encode and copy entire signed capsules to the system clipboard for zero-latency transport between environments.
• ⚡ Two-Pass Consensus & Trajectory OCR: Fully integrated with the SpeedOCR engine to harvest rendered UI states with >98% line recall.

---

🚀 Quick Start: Clipboard Teleportation

Experience the power of instant workspace transport:

1. Seal & Copy Workspace (Machine A)

python3 src/hdar_cli.py seal --workspace src --to-clipboard

Result: Capsule sealed, signed, and copied to system clipboard!

2. Restore Workspace Instantly (Machine B)

python3 src/hdar_cli.py restore --from-clipboard --target /tmp/teleport_target

Result: Byte-exact workspace restored instantly from your clipboard.

3. Verify Security Predicates

python3 src/hdar_cli.py verify --capsule /tmp/enterprise_test.hdar.tar.gz

---

🧮 Theoretical Bounds & Metrics

Character Error Rate (CER)

$$\text{CER} = \frac{S + D + I}{N} = \frac{\sum_{i=1}^{M} \operatorname{lev}(g_i, r_i)}{\sum_{i=1}^{M} |g_i|}$$

Capture Miss Probability ($P_{\text{miss}}$)

$$P_{\text{miss}}(T, f_{\text{effective}}) = \max\left(0, 1 - f_{\text{effective}} \cdot T\right)$$

Merkle Workspace Hashing

$$\mathcal{H}_{\text{root}} = \operatorname{SHA256}\left( \text{CanonicalJSON}\left( { \text{rel_path}_i : \operatorname{SHA256}(\text{Block}i) }{i=1}^K \right) \right)$$

---

🌐 Live Resources

• Interactive Sandbox & Live Verifier: https://overandor.github.io/hdar-canonical/
• Technical Roadmap: ROADMAP.md
• License: Licensed under the MIT License