Home / Articles / CI behaviour policies / Versions / v1.1.0
CI behaviour policies: what changed in v1.1.0
From v1.0.0 (2026-11-16, 2f1456a5b) to v1.1.0 (2026-11-19, eee87768e), paragraph by paragraph.
13 paragraphs added, 0 removed, 2 changed in place, 37 unchanged. About 406 words added and 0 removed. Insertions are marked like this, deletions like this; unchanged runs are folded to one line; figures appear as their file names.
9 unchanged paragraphs, under In short {#in-short}
• Updated 19 November: the job is nine calls; since the 16th the grant has grown to a second organisation and a second DNS zone (three days later).
21 unchanged paragraphs, under A simple job, a large grant {#grant}, Why five hundred repositories? {#scale}, Why we used to get away with it {#excuse}…
Three days later: the whole list, and a wider grant {#update}
On 19 November I went back to the factory's code to write down exactly what one request makes it do. It is a short list, nine calls in all:
• Create the repository in the owner's organisation.
• Push the seed commit from the site template, which includes the site's own Pages workflow.
• Set the repository's settings: default branch dev, and the website in its About box.
• Turn on GitHub Pages, built by GitHub Actions, with a github-pages environment that only dev may deploy from.
• Set the Pages custom domain, after the DNS step.
• Start the site's Pages workflow on dev.
• Add the owner as a collaborator and give the team its access to the new repository.
• Write one CNAME in Route 53, from the site's name to GitHub Pages.
Against that list, the grant has grown since I wrote this piece three days ago. On 17 November the App was also installed on the PrompTugal organisation, on all repositories, so that the factory could make sites there; its own notes say it plainly: "the App's power now covers two organisations". The DNS role now covers a second zone, promptugal.news, still CNAME records only. And the team step needs a permission the App's documented settings do not list: adding a team to a repository needs the organisation permission Members, and the factory's code says so when GitHub refuses. Whether it has been granted is not in the repository.
So the gap is now nine calls against the power to change the settings and workflows of, and delete, every repository in two organisations. Every one of the nine is a small, nameable action. Written as a feature behaviour policy, the factory's job is a page, and each line of it can be enforced by something other than the pipeline's own grant. That is why I think pipelines, and the actions they call, need behaviour policies of their own: the power we hand a pipeline is decided by what the platform's permission model can express, not by what the job needs.
3 unchanged paragraphs, under Why this matters beyond one factory {#why}, What this does not claim {#limits}
• Nothing was exploited. This is a reading of the grants and the workflow, 16 November 2026.2026, and of the factory's code again on 19 November.
4 unchanged paragraphs, under Where this comes from {#sources}
AgentiCCells__Site__Factory, read on 16 November 2026:2026, and again on 19 November at 2f17c90 (build/real.mjs, build/access.mjs, build/io.mjs, docs/github-app.md, docs/aws-oidc.md): docs/github-app.md, docs/aws-oidc.md, docs/SETUP.md, .github/workflows/apply.yml, build/first-pass.mjs, board/FAC-007.md and the merge commits. abp.sgit.ai: the capability grammar, the undo class and the barrier. Earlier on this site: the site factory works, where assist becomes execute, what can your agent reach?, approve at altitude.