Skip to content

ASI Proxy Loop v0.9

kadubon edited this page Jul 1, 2026 · 1 revision

ASI Proxy Loop v0.9

PIC v0.9 adds a bounded ASI-proxy loop for agents. It is not an ASI detector and it is not an execution system. It is a way to turn finite traces into candidate reports, keep residual work visible, and hand repair tasks to CCR or another runtime.

Plain Meaning

The loop is:

trace -> token candidate -> admissibility check -> residuals -> next repair task

The useful output is a JSON report. The report can say what is ready for review and what is still missing. It never grants authority to run a shell command, call a provider, mutate a repository, or operate a physical system.

First Files

Use these files in a source checkout:

  • docs/asi-proxy-loop.md
  • docs/agent-loop-protocol.md
  • examples/asi_proxy_loop_bundle/target.json
  • examples/asi_proxy_loop_bundle/pic_extraction_pipeline.example.json
  • examples/asi_proxy_loop_bundle/pic_token_admissibility.example.json
  • examples/asi_proxy_loop_bundle/capital_witnesses.jsonl

First Commands

pic token extract-pipeline --trace trace.json --compact
pic token admissibility --token token.json --compact
pic performance report
pic cache status

Safety Boundary

  • accepted=true is not settlement.
  • token_extraction is not token execution.
  • Token admissibility is not capital admission.
  • operation_ready=true is not provider dispatch.
  • physical_ready=true is not physical outcome proof.
  • safe_commands are review hints, not authority.

Search Terms

ASI-proxy loop, CARA loop, agent loop protocol, token extraction, token admissibility, residual repair, CCR loop import, observation residual, phase acceleration interval, pic token extract-pipeline, pic token admissibility, settled=false.

Clone this wiki locally