Skip to content

For the complete documentation index, see llms.txt.

Promptless for Agent Instructions

Promptless Instruction Governance (PIG) helps teams maintain the instructions their AI agents use. Keep shared skills, rules, commands, and tool configuration in a versioned Instruction Hub. Publish them as plugins, then use evidence from real sessions to improve them.

You can start with instruction publishing on its own. When you are ready to learn from agent sessions, deploy the trace analyzer in your infrastructure and enroll the hosts that should send it traces.

Instruction HubPublish shared instructions
Published pluginsthen
AI agentUse instructions in real work
Session tracesthen
Trace analyzerFind instruction problems
Findingsthen
Promptless DashboardReview evidence and improvements
Reviewed instruction updates → Instruction Hub
Publish instructions on their own, or add trace analysis to learn from sessions. Review proposed improvements and publish accepted changes back through your hub.

See How it works for the full architecture, ownership boundaries, and data flow.

  • Copies of the same skill drift apart across repositories. Keep one reviewed source in an Instruction Hub, with clear ownership and a release history.
  • Your writers and developers use different agents. The pig toolchain builds agent-specific plugins from the same hub. Authors maintain one source, and each agent receives a format it supports.
  • Agents repeat a mistake, and no one knows which instruction causes it. The trace analyzer examines real sessions for missing, stale, conflicting, or ineffective instructions. It records each problem as a finding with supporting evidence.
  • Instruction changes must not reach your team without review. A finding can lead to a proposed pull request against your hub. Your team reviews it and publishes accepted changes through its usual Git workflow.

For example, Acme’s writers and developers share a review-docs skill. A finding shows that an older release of the skill omits prerequisite checks, so agents approve guides with missing setup steps. A reviewer checks the cited session evidence, merges a correction, and publishes a new release for the team to install.

The hub guides are for technical writers and developers. The deployment guides cover the Kubernetes, identity, storage, and operational decisions that DevOps and GitOps teams manage. You can begin authoring instructions without the deployment details.

To hand setup to Claude Code, Codex, or another coding agent, start with Set up PIG with a coding agent. It has a prompt for the whole setup, and each setup guide has a prompt for its own phase.

An Instruction Hub needs a Git repository, the pig toolchain, a publishing pipeline, and an agent that can install the output. It does not need a trace analyzer, database, bucket, or model-provider account. New hubs have managed trace ingestion disabled by default.

Use this path to establish your source of truth, test the authoring workflow, and get useful instructions into your team’s agents. Create a hub or migrate existing instructions.

For trace analysis, deploy the analyzer and connect PostgreSQL and your cloud’s object storage. Configure a model provider, select the instruction repositories to analyze, and then enable host enrollment. Raw traces stay in your storage. The configured model provider receives analysis input, and Promptless receives trace metadata and findings that can include session-derived detail.

The trace analyzer updates automatically to stable releases by default. You can pause updates or pin a release. See Manage updates and recovery.

Before you enable collection, read the trust and data model with your deployment team.

Author, publish, and install shared instructions without deploying a trace analyzer.

Deploy the analyzer, enroll hosts, and act on the findings it produces.