[gustavonline]

Factory's portable guidance did not yet express the selected two-reviewer default, project context bootstrap or the current Lab's actual isolation limits. This exports the corresponding Notion originals at the recorded revision and updates README, architecture and qualification guidance.

• Default to two fresh independent reviewers: objective/domain evidence and qualitative intent/outcome. Allow project/task overrides within mandate and stronger rules.
• Clarify Planning Agent → Foundation → the project's own Context Home → AgentOps, Factory-only Notion read/update/insert for the lead, and selected files/model-only access for workers.
• Make the human PR decision handoff explicit (Gør det tydeligt, hvornår en PR kræver menneskelig godkendelse #88): delegation can cover ordinary delivery; silence, refusal, changed candidates and new material exposure require the affected work to wait or be reassessed. Require real before/after captures for meaningful visible results (prs should always include screenshots or screenrecordings depending on what the it is about #95), with proportionate exceptions.
• Label prepared sibling VMs as a qualification layout. The unqualified target remains on-demand isolated sandboxes per agent, managed inside AgentOps' environment, with minimal physical-host actions and no additional coordinator LLM.

Validation

Exact reviewed candidate: bfe9d987d28fa60b59f4f0f76290e1a53972caa5.
Both new independent Claude Opus 5.5 High, Fast off source reviews ACCEPT, no blockers, after the previous pair's findings were repaired. Native T3 thread/run model selections were read back separately by the operator; the reviewers' environment/settings observations alone were not used as proof.

Both reviewers independently verified the 96-member packet manifest, complete Git bundle, exact diffs, source/export correspondence, unchanged inputs, all 11 supporting-file hashes and preserved portable links. Each ran the existing checker (9 skills, 0 errors), all 16 tests and staging/link validation on Z13. No scripts, tests or workflows changed.

The four #88 decision packets are synthetic source-behavior rehearsals, not four real PR demonstrations. The reviewers accepted them for the scoped method change. No screenshots were invented or counted as code execution evidence.

Native observations and limits

The Factory-only Notion create/read/update/readback/trash and excluded-page 404 checks ran on T3 0.1.2 before the 3796 cutover, not as a second test on Nightly. The later native author → two reviewers → repair → two new reviewers flow completed on official Nightly 3796. An approval needed operator detection/steering; a stale approval remained unanswered. Remote isolation:worktree was refused. A fresh read-only worker probe was denied access to the lead.

Private operator installation records additionally support the official archive checksum, migration counts, native pin/auto-settle setting, worker recovery copies, expected recovered environment and completion wakes. Those operator observations were not reproduced by the source reviewers. Two idle crash cycles and scratch restoration do not qualify reboot, active-write recovery or full native restore/resume.

Guest sudo/Docker makes controls inside that guest advisory. Connector flag efficacy under account OAuth and worker WebSearch/WebFetch/RemoteTrigger/GitHub-git routes remain unqualified. A diagnostic gVisor component preflight is not a model/T3 recipe, VM-equivalence result or benchmark.

No unattended intake, automatic provisioning, new backend, host upgrade or other-factory adoption is enabled by this PR. #32, #34 and #80 remain open. Version 0.1.26 is a portable distribution; installation is recorded separately.

Canonical source: the Notion Factory Skills library snapshot d8cca3c0a7a04b5f9f77962ae47109283bafa1234f5f61a1298185be4d91cd7f; all required supporting files were exported deliberately. YAML original 0.3 is unchanged, SHA256 98f48a4287d9b4fd14c39780adc70c5932af2185fd3546b22a5fab97f8f7647f, code-block edit 28 October 08:09Z. Git remains a reviewed export, not a second editable original.

Delivery follows the owner's explicit instruction to complete this scoped work and retains ordinary branch protection. Credits in the final merge must preserve the human author and actual Codex / Claude contributions.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-28T10:56:29.386116Z	bfe9d98	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.