Good fit if
- Engineers and QA leads evaluating agent quality
- Teams gating releases on written regression checks
- Evaluator agents applying anchored rubrics
Quality & Evaluation
Wave 2 · launch-sequenced
Agent Evaluation Harness Suite
Ready-to-run evaluation harnesses for agent behavior - task suites, anchored graders, and regression gates that tell you whether an agent change made things better or worse before it ships.
v1.0.0-rc1
Medium risk
20-min quick start
Select a one-time tier below
View agent record
Built for: agent-engineers · qa-leads · evaluator-agents · multi-agent-operators ·
agent://products.json
200 OK
{
"id": "agent-evaluation-harness-suite",
"version": "1.0.0-rc1",
"price_usd": { "lite":39, "std":99, "prem":169 },
"risk_level": "medium",
"reversible": true,
"checkout": { "enabled": true },
"purchasable": true
}
See how it fits your agent stack
AI output
to be judged
Agent Evaluation Harness Suite
AEL-6 method
Task-Suites/
Graders/
Regression-Gates/
Run-Records/
Scored output
pass / revise / fail
Human approval gate
Preview-before-apply
Rollback path
Discover
read the record
Evaluate
Medium risk · 6 systems
Install
new Task-Suites/
Preview
diff before apply
Approve
human review
Audit & rollback
ECS-100 audit
Non-destructive by rule — new-folder install, never overwrite, human-approval gates for risky steps, rollback included.
agent-evaluation-harness-suite/
Task-Suites/
Graders/
Regression-Gates/
Run-Records/
Worked-Evaluations/
README.md
AGENT-INSTRUCTIONS.md
Integration/ROLLBACK-GUIDE.md
Format & setup
FormatMarkdown + JSON file kit (plain text, tool-agnostic)
Skill levelIntermediate; beginner-usable via quick start
Quick start20 min
Full setup120 min
AEL-6 framework, task-suite/run-record/gate-config JSON schemas, six anchored rubric families, written regression gates, ECS-100 coverage scorecard, worked regression cycle + three-agent-type walkthrough, 16-doc integration pack, 5 upgrade modules, 10-doc differentiation folder
# Non-destructive lifecycle — nothing runs without preview + approval
$ agent read ./agent-evaluation-harness-suite/manifest.json # metadata, risk, compatibility
$ agent install --new-folder ./agent-evaluation-harness-suite # never overwrites existing files
$ agent apply --preview # preview-before-apply diff
$ agent verify ./agent-evaluation-harness-suite --scorecard ECS-100 # Task-Suites contracts
⚠ gated steps require named human approval before execution
$ agent rollback ./agent-evaluation-harness-suite # Delete the install folder (complete remo
✓ folder removed — nothing else touched
{
"required_permissions": ["write access to one new buyer-chosen folder"],
"integration_modes": ["standalone", "drop-in-upgrade", "advanced-integration", ],
"compatible_systems": [
"Claude (Projects, Skills, Claude Code)",
"ChatGPT (Custom GPTs, Projects)",
"any LLM chat assistant",
"MCP-based agent stacks",
],
"license_type": "single-operator",
"checkout": { "enabled": true, "status": "polar_active" }
}
The problem it solves
Agent changes ship on vibes: no task suites, no anchored graders, no regression gate, no comparable run records.
Use cases
gate a prompt change before shipping
swap a model with per-dimension evidence
vet a sub-agent before widening its toolset
prove injection defenses hold after a change
produce a deployment-readiness evidence package
Safety & permissions
Non-destructive by rule
Plain-text kit; evaluates and records, executes nothing, modifies no agent under test, never auto-ships a change. Handles no credentials or wallets. Defensive/quality use only.
Required
write access to one new buyer-chosen folder
Rollback
Delete the install folder (complete removal). Deeper integrations: Integrations/ROLLBACK-GUIDE.md.
For an evaluating agent
File kit for evaluating agent behavior on AEL-6 (Define/Suite/Run/Grade/Gate/Record): task suites, six anchored rubric families, written regression gates, run records that bisect regressions, ECS-100. Evaluates and records only; never auto-ships. Builds on P01/P02; feeds P16. Non-destructive install, delete-folder rollback, no credentials. Wave 2, built Production: live and purchasable through verified Polar checkout; prices owner-confirmed 2026-07-03 (still production); verified Polar checkout is active on this page.
Buyer decision
Is Agent Evaluation Harness Suite right for me?
Problem addressed: Agent changes ship on vibes: no task suites, no anchored graders, no regression gate, no comparable run records.
Probably not a fit if
- Buyers expecting an eval runner or benchmark software
- Teams with no agent to evaluate
- Anyone expecting hosted dashboards or leaderboards
Included
Task-Suites/, Graders/ (7), Regression-Gates/, Run-Records/, Worked-Evaluations/ folders plus the standard architecture.
Not included
- No software, app, executable, or code that runs by itself
- No hosted service, account, login, or cloud dashboard
- No live integrations: Zapier/Make items are written blueprint descriptions, not built Zaps or scenarios
- No API access, API keys, credentials, or third-party subscriptions
| Tier | Price | Canonical depth | License |
|---|---|---|---|
| Standard | $99 USD | Full product: all templates, harnesses, SOPs, integration pack, and samples. | Single-Operator |
| Premium | $169 USD | Standard plus all 5 upgrade modules (Basic, Pro, Agent, Automation, Safety) and priority support. | Single-Operator |
Bundle membership
Compatibility and limits
You run evaluations with your own models/agents and record results in the provided schemas - no runner code is included 'Harness Suite' is a document+schema suite, not an application
Three one-time tiers — verified Polar checkout active
Lite
$39
Core templates & quick-start path. Best for testing the system manually first.
Buy Lite — one-timeStandardpopular
$99
Full product: all templates, harnesses, SOPs, integration pack, and samples.
Buy Standard — one-timePremium
$169
Standard plus all 5 upgrade modules (Basic, Pro, Agent, Automation, Safety) and priority support.
Buy Premium — one-timeEach tier uses its verified Polar checkout. See full pricing and bundles.
Related products
Prompt Harness Builder Kit
Prompt & Content Systems
AI Output Quality-Control Scorecard System
Quality & Evaluation
Agent Safety Case Builder
Safety & Security
Agent Error Recovery Playbook
Reliability & Operations
In bundles: Wave 2 Agent Trust & Safety Stack ·
Decision guides involving this product
- Agent Evaluation vs Output Quality ControlDecide whether you need task-level agent regression evidence, output-level review rubrics, or both layers together.
- Which SignalHarness Bundle Fits My Use Case?Compare the nine canonical bundles by buyer profile, workflow coverage, included products, and when individual products are a better fit.
Related categories
- AI Evaluation & Quality ControlRubrics, task suites, regression gates, safety evidence, and audit records for judging AI and agent behavior.
Available in canonical bundles
These bundle memberships come directly from the canonical bundle catalog. Compare the exact contents before purchase.