Riposte

From intent
to evidence.

Design specification

Riposte is a conversation with an observable state machine beneath it. Every transition has an owner, a record, and a failure mode. This page describes the designed behavior of a product that has not shipped yet.

Objective

You state the outcome and scope. Repository, target, provider, and tool content remain untrusted data.

conversation.message

Plan

The agent proposes visible steps and can revise them. A plan never grants permission.

plan.versioned

Policy

Canonical resources, effects, risk, policy, and approval bind the exact action.

decision + grant

Execution

The native boundary enforces filesystem, process, network, time, and budget constraints.

execution.receipt

Evidence

Raw and normalized observations become content-addressed evidence with provenance.

evidence.record

Finding

Claims cite evidence, preserve counterevidence, and expose coverage and confidence.

finding.projection

CHAT-FIRST, NOT CHAT-ONLY

One calm surface.
Deep inspection on demand.

Running riposte is designed to open the chat TUI directly in your workspace. Everything below is part of that one surface, not a separate app.

Command palette

Cases, settings, exports, tools, and diagnostics without memorizing modes.

Expandable activity

See the useful result first; reveal raw output, timing, scope, and provenance when needed.

Resumable cases

Close the terminal, reconnect to a run, review history, and continue from durable state.

Headless when required

The same contracts are designed to support CI and automation with machine-readable output and no fake TUI.

02 / THE RECORD

What a case remembers.

Security work does not end when the terminal closes. A Riposte case is designed as a durable unit of work whose parts stay connected: what you asked for, what was proposed, what was allowed, what actually ran, and what was learned.

case
Durable purpose, scope, owner, and sensitivity for a piece of work.
plan
Versioned proposed steps; revisions preserve what was proposed before.
action
The normalized intended effect, bound to policy and approval before execution.
receipt
The durable outcome of an execution: observed effects, limits, and uncertainty.
evidence
Acquired or derived material with integrity, provenance, and classification.
finding
An evaluated claim that cites evidence and keeps confidence and gaps visible.
Why it matters

When a tool disconnects mid-run or a session crashes, the honest state is often "unknown". Riposte is designed to record unknown outcomes as unknown instead of guessing success or failure.

03 / EXTENSIBILITY

Extensible, behind the same boundary.Designed

External functionality enters Riposte as typed capabilities, not free-form shell access. Security disciplines are packaged as packs; tools declare their inputs, effects, risk, and evidence types; MCP servers and plugins are mediated by the same policy and containment rules as built-in tools.

  • Capabilities declare effects and risk before anything runs.
  • Packs bundle the tools, knowledge, and workflows of one security discipline.
  • Plugins and MCP integrations receive scoped, revocable trust, never ambient trust.

Public extension contracts, signing, and a review process are part of the design and will be documented before third-party extensions are supported.

Not a marketplace

No extension marketplace will launch before trust tiers, review, revocation, and maintenance states are operational.

04 / AVAILABILITY

Current state, stated plainly.

Everything on this page is design intent backed by versioned specifications, with a private vertical slice in active development. There is no public build, no package to install, and no support commitment yet.

When that changes, this site will show the exact version, platforms, verification artifacts, and a tested install path. Until then, the honest next step is to read the security model or follow the project.