TechTide Harness Kit
Trust contracts for agent skills expressed as JSON Schema: risk tiers, tool permissions, deny-by-default egress, approval gates, and prompt-injection fixtures behind a wall of CI validation gates.
Proof ledger
- Classification
- Built artifact
- Repository
- TechTideOhio/techtide-harness-kit
- Language
- Python
- Licence
- Apache-2.0
- Status
- Maintained
- Tests
- 3 recorded gates
- Last verified
- 2026-07-27
What it is
The Harness Kit turns the informal rules people write in prompt files into machine-checkable contracts. A skill declares its risk tier, the tools it may touch, and the network egress it needs. CI then rejects any skill whose declaration does not validate, and prompt-injection fixtures assert that a skill refuses instructions it should refuse.
Architecture
- JSON Schema contracts define risk tiers, tool permission sets, and egress allowances per skill.
- Deny-by-default egress: a skill reaches nothing it has not declared.
- Approval gates attach to tiers rather than to individual call sites, so escalation is uniform.
- Prompt-injection fixtures live beside the contracts and run as tests.
- Provider scope regression checks guard against a skill silently widening its permissions across releases.
Tests and gates
- Nine GitHub Actions workflows: ci, codeql, docs-quality, install-paths-smoke, packed-artifact-smoke, provider-scope-regression, apply-ruleset, release, and OpenSSF scorecard.
- The README describes roughly 28 CI validation gates across those workflows.
- CONTROL-MAPPING.md, DATA-HANDLING.md, TRUST.md, EVALS.md, and PROMPT-INJECTION.md document what each gate is asserting.
Limits
- The kit validates declarations and fixtures. It cannot prove a model will never be jailbroken, only that declared boundaries are enforced by tooling.
- Contracts are only as tight as the author makes them; a permissive tier is still permissive.
- Coverage is aimed at Claude Code, Codex, Cursor, and Kiro style skill packages, not at every agent runtime.
Source artifacts
- README →github.com/TechTideOhio/techtide-harness-kit#readme
- Trust contracts →github.com/TechTideOhio/techtide-harness-kit/blob/master/TRUST.md
- Prompt-injection fixtures →github.com/TechTideOhio/techtide-harness-kit/blob/master/PROMPT-INJECTION.md
- Control mapping →github.com/TechTideOhio/techtide-harness-kit/blob/master/CONTROL-MAPPING.md
- Evals →github.com/TechTideOhio/techtide-harness-kit/blob/master/EVALS.md
Patterns applied
Acceptance tests for agent work →
Define done before the agent starts.
Evals that gate releases →
A golden suite in CI, not a spreadsheet.
MCP permission scoping →
Deny by default, declare every tool.
Observability for agent runs →
One trace ID from request to receipt.
Cost controls that actually stop spend →
Ceilings enforced before the call, not after the invoice.
Approval gates that suspend the run →
Nothing downstream executes before a human approves.
Related proof
ClawKeeper →
Governed Claude agent control plane. A deterministic policy engine runs before every LLM call, autonomous spend defaults to zero, and the audit log is enforced by the database.
CipherClaw →
Deterministic multi-agent debugger with zero LLM calls in the analysis path: causal graphs, weighted root-cause scoring, and failure prediction.
Swarm 357 →
357 Claude agent roles across six business layers, with portable memory, enforced cost caps, a bash policy gate, and durable checkpoints.
FintheFinder →
Deep-research agent whose runs suspend until a human approves, with a deterministic citation audit that keeps AI out of the verification path.
This page is implementation proof only. Engagement scope and pricing live on TechTide AI.
AI Production Systems at TechTide AI→© 2026 Alex Cinovoj · TechTide AI · Columbus, OH