Promptless for Agent Instructions
Promptless for Agent Instructions governs the skills, subagents, hooks, commands, rules, and MCP configs your agents run on: it collects evidence from real agent sessions, finds where those instructions fail, and opens reviewed pull requests that fix them.
The instruction-drift problem
Section titled “The instruction-drift problem”As teams move operational knowledge into agent-read files — skills, subagent definitions, hooks, commands, rules, MCP configs, and AGENTS.md or CLAUDE.md context — that knowledge becomes infrastructure. And like any infrastructure, it drifts. The result is instruction debt, sometimes called skill slop: a growing pile of instructions that quietly make your agents worse.
Instructions drift in specific, recurring ways:
- Stale — the procedure changed, but the instruction still describes the old one.
- Missing — a new workflow has no instruction at all.
- Contradictory — two skills tell an agent to do opposite things.
- Dead — an instruction references a command, tool, or path that no longer exists.
- Unowned — no one is responsible for keeping the instruction correct.
- Untestable — there is no way to tell whether the instruction still works.
- Unenforced — the instruction exists, but agents ignore it.
- Undistributed — the canonical instruction never reaches the agent that needs it.
- Non-portable — the instruction assumes one harness or format and breaks elsewhere.
- Risky — the instruction changes permissions, autonomy, or escalation behavior.
Left unmanaged, these failures compound. Agents thrash, repeat corrected mistakes, and take unsafe actions, and the people who own the instructions have no evidence to tell them which instruction is at fault.
Who it’s for
Section titled “Who it’s for”Promptless for Agent Instructions is built for the people who own an agent fleet across a company: AI-governance leads, platform and AI-engineering teams, and engineering leadership. The value grows with the size of the fleet, so it matters most once a team runs dozens of skills across many agents and hosts.
If you already run security and observability tooling — the kind of continuous, evidence-driven scanning that Semgrep, Wiz, Snyk, or Vanta provide for code and infrastructure — this is the same posture applied to agent instructions. You get findings backed by real session evidence, not one-off audits.
How the loop works
Section titled “How the loop works”You author and organize your instructions in an Instruction Hub, a versioned Git repository. The pig toolchain compiles that hub into target-native plugin packages, and your fleet installs them. On each host, a runtime enrolls with a worker you deploy and streams native session traces to it. Inside the worker, the Friction Analyzer studies those sessions and records findings, each backed by append-only evidence. Findings become GitHub issues, high-confidence findings open remediation pull requests against your hub, and a human reviews and merges the fix — which recompiles and redistributes to close the loop.
For the full end-to-end path, see How it works.