Skip to content

test(headless): stage scoped MCP agent qualification seam - #337

Draft
0thernet wants to merge 2 commits into
mainfrom
codex/headless-agent-qualification-20261015
Draft

0thernet wants to merge 2 commits into
mainfrom
codex/headless-agent-qualification-20261015

Conversation

@0thernet

@0thernet 0thernet commented Oct 15, 2026 •

Copy link
Copy Markdown
Member

Draft fixture seam, NOT a two-agent qualification pass or a release gate.

Current candidate 5be707998c2f21b48f0a41adc31ea1957678caa1 (tree b717b1bbf44753c1a411bb4eb2fcfebacecda8b6) modifies only .github/scripts/headless_agent_qualification.py and its tests. It adds bounded local retry/outbox and peer-inbox checks, separate finite-grant issuance contracts, fail-closed MCP validation, and private grant-file cleanup. These are local fixture contracts; machine IDs are caller declarations, not independent placement proof.

Agent-decided, owner may revisit: pin the two-file incremental candidate on this existing draft PR to exercise exact-head Linux Required, because the local sandbox denies AF_UNIX bind with EPERM. Keep draft while the fixture and live evidence remain incomplete. Before push, five focused suites ran 108 tests (107 passed; the opt-in real-process test skipped); git diff --check and the staged check passed. CI's existing workflow does not set HEADLESS_AGENT_TEST_BINARY, so its Required check alone cannot prove the skipped real-process test or the two-machine journey.

Before marking ready or requesting squash auto-merge: run the opt-in exact-source test on a socket-capable runner, create two independently scoped agents on distinct disposable VMs against a pinned host, exercise signed send/read, typed refusals, offline catch-up, same-home restart and duplicate-free retry, and verify sanitized source/tree/lock/binary/run/attempt/machine/cleanup receipts through root-owned workflow/result integration. Recheck exact-head Required/security, review threads and conflicts. No grant file, bootstrap secret, message body, home, live two-agent receipt or deployment is published. PR #303's custody diagnosis remains separate.

@vercel

vercel Bot commented Oct 15, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
valhalla Ready Ready Preview Oct 15, 2026 5:27pm UTC

Request Review

Add bounded local retry and peer inbox scans, two disjoint finite-grant issuance contracts, fail-closed response validation, and grant-file cleanup after confirmed revoke. Deterministic tests cover these seams; the opt-in real-process test and independently verified two-machine journey remain outstanding. This is not a qualification pass.

This branch was successfully deployed

1 active deployment
Preview — 5be70799 Deployed Oct 15, 2026 by vercel[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant