Repository navigation
Add opt-in Pi durable harness with bounded sandbox recovery - #29
Merged
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
ESF needs an opt-in Pi harness that can recover its internal agent loop inside a surviving Cube microVM while preserving factory verification, evidence ownership, and Temporal's single-attempt agent policy.
Changes
type = "pi"with fixed JSON stdin transport, startup model/endpoint validation, and digest verification of the runner and Node runtime.Verification
Executed project code and tests inside local CubeSandbox microVMs:
npm ci --ignore-scripts,npm run build,node export-catalog.mjs,npm test: 16 runner regressions passed; rebuilding reproduced the committed runner digest.npm audit --audit-level=high: zero vulnerabilities.internal/agentharness,internal/factory, andcmd/factorysuites passed with the race detector. Test binaries were compiled withgo test -c -raceand executed inside Cube; final model/pricing changes also passed focused regressions.factory agents verify, candidate inventory verification, offline installation with the template's Python 3.10, and modified-runner digest rejection passed.go vet ./internal/agentharness ./internal/factory ./cmd/factoryandgit diff --checkpassed.Risks
Pi 1.0.3 exposes an experimental API. Recovery covers confirmed engine death in a surviving sandbox; active shell execution, surviving processes, uncertain inspection, lost state, and supervisor failure remain failed runs. It does not guarantee exactly-once tool effects.
This PR remains a draft pending model-backed qualification. The local gateway was unreachable; an HTTPS provider/model/credential profile and approved custom-model limits are still required. Real-model acceptance and same-task/model comparisons with OpenCode and Unreal remain pending. Deterministic results do not establish production readiness.