Tethermark
Open AI security audit harness for agentic repositories and local AI systems, focused on evidence-grounded static audits, supervisor QA, exports, and gated runtime validation.
Problem / Scope
Tethermark is an AI security audit harness for agentic repositories and local AI systems. The near-term goal is to make security review repeatable enough to share: deterministic evidence collection, static analysis, structured findings, scoring, exports, and clear disclosures around what is validated versus still in development.
Architecture
- Static audit pipeline for repository posture, dependency signals, secret exposure, policy files, and agent/tool boundary indicators
- Supervisor review layer for findings, severity, evidence citations, and score explanations
- Export path for Markdown, JSON, and website-ready artifacts
- Runtime doctor and sandbox boundary work for local execution checks where a container runtime is available
- Planned publication path into AISecurityBase for ranked research and newsletter content
Key Workflows / What It Proves
- Run a repository audit and produce evidence-backed findings
- Separate static readiness from runtime validation so claims stay bounded
- Use repeatable fixtures and example repositories to create screenshots, reports, and launch content
- Convert audit output into public-facing research assets without overstating assurance
Limitations
- Runtime validation is still gated by local container backend availability and additional test coverage
- Newer features such as chat assistant, self-learning, and benchmark flows need more validation before launch claims
- Public examples should be rerun before posting so screenshots and findings reflect the current code state
Evidence Pack
E-TETHERMARK-001
Static audit run snapshots
Validated local fixture scans showing scored findings, static baseline behavior, and evidence-grounded output.
E-TETHERMARK-002
Runtime readiness note
Current runtime boundary and dependency status, including Docker/Podman/gVisor requirements and known blockers.
E-TETHERMARK-003
Launch content checklist
Post-by-post readiness tasks for screenshots, repository examples, feature demos, and known-gap disclosures.