AgentOps Plugins / Governance

Harness Governance Plugin

Package harness reflection, proposal drafting, and regression evaluation into one governed harness-improvement plugin with explicit human gates.

Not verifiedComponent bundleRequirements & limits

Inputs

The released example shows these fields. Check the guide for which are required.

  • run_id
  • observation
  • proposed_change
  • regression_result

Outputs

  • A bounded decision record for a proposed harness change

Before you begin

  • Node.js 22 for the packaged sample. Harness adoption requires a separate human decision.

Try this first

Inspect the packaged offline sample before running it. After file installation, use the stated runtime from the installed package directory.

Limits for this step
  • Harness adoption remains human-approved; the packaged Markdown hook is advisory on runtimes without a native enforcement mapping.
  • Packaged programs have independent setup and verification requirements. Reading or copying this procedure does not verify their execution.
Inspect offline sampleHide offline sampleSynthetic example

Runtime · Node.js 22. Run from the installed package directory after file installation.

node scripts/first_use.mjs examples/first-use-input.json

Expected result · A bounded decision record on standard output. It does not authorize adoption.

This is the release’s documented sample, separate from evidence of operation in your environment.

Sample input
{
  "run_id": "demo-run-001",
  "observation": "Repeated tool retries consumed the context budget.",
  "proposed_change": "Add a bounded retry policy to the harness guidance.",
  "regression_result": "pass"
}

Summary sources · Catalogue metadata, plugin.yaml, README.md, examples/first-use-input.json · v0.2.0 · Release package

View package contentsHide package contentsv0.2.0

Bundled skills

Packaged hooks

require-human-approval-for-harness-adoption

require-human-approval-for-harness-adoption

Before any harness proposal is accepted:

  1. Confirm the proposal stays within harness-owned files.
  2. Require benchmark and red-team evaluation results.
  3. Do not auto-apply accepted proposals.
  4. Route the final adoption decision to a human reviewer.

Install plugin files

These commands install package files and generate runtime configuration where supported. Confirm discovery in your agent app, then check configuration, authentication and a first useful operation.

Install (Codex)

./bin/skills-hub install --module plugins --entry agentops/[email protected] --runtime codex

Install (Claude)

./bin/skills-hub install --module plugins --entry agentops/[email protected] --runtime claude

Install (Generic)

./bin/skills-hub install --module plugins --entry agentops/[email protected] --runtime generic --target ./my-agent/plugins

Bundled skills, agents and tools are placed in the selected runtime directories. Hooks stay inside the plugin directory. Codex and Claude targets also receive a generated plugin manifest.

No security review is declared for this release. Review bundled components, required secrets and approval rules before live use.

  • Approval rules: human-approval-for-harness-update

How You Can Use This

Not verified

Resource form: Component bundle

Current operational evidence for your environment is not shown here. Check the implementation, installation requirements and limits before use.

First run: Run `node scripts/first_use.mjs examples/first-use-input.json` after installation to produce a governed proposal decision.

Current limits

  • Harness adoption remains human-approved; the packaged Markdown hook is advisory on runtimes without a native enforcement mapping.

Packaged executable helpers

Each program has its own setup requirements and operational evidence. Reading or copying the instructions does not verify these programs.

scripts/first_use.mjs · Not verified · Local program

First run: node scripts/first_use.mjs examples/first-use-input.json

  • Requires a declared Node.js 22 execution capability; it never authorizes harness adoption.
What installation and verification mean
File installation
Places package files and generated configuration in the selected directories.
Native discovery
Your agent app finds and accepts the instructions, tools or plugin registrations.
Mock execution
Exercises sample fixtures. Connected data and provider effects need a separate check.
Connected operation
Requires current configuration, provider authentication, scoped permissions and a successful operation in your intended environment.

Declared compatibility

Compatibility unknown: versions and capabilities have not been declared.

Classification: inferred

Technical details and release files

Review and lifecycle

Review maturity: Experimental

Security review declaration: Not declared

Lifecycle: Active

These release declarations are separate from evidence that a workflow operates in your environment.

Metadata

ID: agentops/harness-governance-plugin

Tags: harness, governance, evaluation, reflection

Latest: 0.2.0

Runtimes: codex, claude, generic

Requirements

Secrets: 0

Approvals: 1

Approval rules: human-approval-for-harness-update