@pi-harness/plugin-anchored-standard

Anchored Standard — Audit agent trajectories against an explicit tool-call allowlist and flag activity outside the anchored run.

Install

npm install --save-exact @pi-harness/plugin-anchored-standard

Enable

Add the entry to the Cordis profile the harness starts from:

- id: anchored-standard
  name: "@pi-harness/plugin-anchored-standard"
  config: {}

The Pi Harness plugin marketplace installs and enables this package for you; the steps above are the manual equivalent.

Audit semantics

trajectory_anchor_check {} returns a complete JSON report in both model-visible text and UI details: status, auditOnly: true, scope: "since-plugin-load", event count, tool-call count, configured budget, allowed tools, and violation codes/messages.

• This is a read-only audit, not enforcement: disallowed tools and over-budget calls still execute.
• Events and violations accumulate across sessions for the plugin's lifetime. Only the first finding of each of the five violation codes is retained. Reloading the plugin clears this history.
• toolCalls counts the current or last run and resets on agent_start; a historical violated status does not mean the current run introduced another violation.
• maxToolCalls defaults to 64 (bounded to 1–512); an empty allowedTools list allows any tool. The allowlist contains at most 512 names of 128 UTF-16 code units each.
• Diagnostic tool names are clipped to 128 code units with an ellipsis; allowlist matching uses the full original name. The complete serialized UTF-8 report stays within 512 KiB, including worst-case JSON escaping.