Panda Suite Docs
v1.9.2GitHub ↗
Panda Suite / Panda Invaris

Panda Invaris

Examples use standalone Codex names. For a Codex plugin, use $panda-suite:panda-invaris; for a Claude Code plugin, use /panda-suite:panda-invaris. See all invocation formats.

Panda Invaris helps you inspect, author, run, and interpret Invaris harnesses: configured project checks, guardrail challenges, mutations, and campaigns. Use it when you need evidence that your project's checks run, or that they actually detect a specific deliberate change.

At a glance

Topic Summary
Use it for Explaining a harness, running checks, testing whether a guard detects a violation, testing whether tests catch a code change
Not for General test design (Test), agent or skill evaluation (Evaris)
Requires An installed, compatible Invaris CLI to execute anything; inspection works without it
Edits by default None; authoring or running needs your request
Key rule A passing check or baseline does not prove a guard detects violations

Actions

Action What it answers
check Do the configured guards pass?
run Do the steps and guards pass?
challenge Does a guard detect a specified violation?
mutate Do the selected tests detect a specified code change?
campaign What is the aggregate result of defined experiments under a policy?

The skill checks the installed CLI's version and usage first and does not assume a newer command or schema is supported.

How it works

  1. Establish outcome and authorization: explain only, author or edit a harness, or run specific checks. The harness, its commands, paths, policies, and side effects are inspected before anything runs. Commands run with your permissions; temporary workspaces do not isolate external services.
  2. Check the CLI and matching documentation and schema. If Invaris is missing, the prerequisite is described and inspection continues. init can create an empty harness, which is not verification by itself.
  3. Choose and run the action that answers the request, reporting the actual command, harness, version, exit status, and evidence.
  4. Interpret the native report without flattening it: strong structured proof versus legacy marker evidence; PROVEN, UNPROVEN, INCONCLUSIVE; KILLED, SURVIVED, INVALID, INCONCLUSIVE. Exit 2, missing tools, timeout, refusal, or invalid evidence is never a pass. Mixed or incomplete campaign results are preserved.
  5. Evaris integration, if requested, follows Evaris's current compatibility contract. An Invaris ACCEPT never proves task completion.

Examples

Explain a harness

$panda-invaris Inspect .invaris/invaris.json and explain which guard checks
access control. Run the authorized challenge and report its actual evidence.

Explain only

$panda-invaris Explain what each guard in .invaris/invaris.json does and what
side effects running it could have. Do not run anything.

Run the checks

$panda-invaris Run check with the project harness and summarize each guard's
result with the native evidence path.

Does the guard catch a violation?

$panda-invaris Challenge the "no secrets in config" guard with a synthetic fake
API key in a temporary copy. Report PROVEN, UNPROVEN or INCONCLUSIVE.

Do the tests catch a code change?

$panda-invaris Mutate src/auth/owner.py by removing the owner comparison and run
the auth tests. Report KILLED or SURVIVED with evidence.

A SURVIVED mutation shows a test gap worth handing to Test.

Author a harness

$panda-invaris Create a harness for this Node project with guards for lint,
type check and unit tests, using the commands in package.json. Validate it with
doctor if this Invaris version supports it.

Campaign

$panda-invaris Run the campaign in .invaris/campaign.json and summarize results,
keeping incomplete and ineligible experiments visible.

Missing CLI

$panda-invaris Run the harness checks.

If Invaris is not installed, expect the missing prerequisite to be named and no invented results. Nothing is installed without authorization.

Boundaries

Untrusted commands are not run. Existing configuration and your changes are preserved. Reports are never changed and guards are never weakened to produce a passing result.

  • Evaris for agent and skill evaluations
  • Security audits harness evidence paths in skills mode

Skill instructions

Read the full Panda Invaris instructions in the repository.