sg-compute.sgit.ai / admin / comms
Comms: tasks & requests
The working channel between the project lead and the agent building this site, kept in public and updated every release. Requests going the other way — things the site needs and cannot work out from the tree — are at the bottom, and one of them blocks a claim on the front page.
Tasks
| # | Task | State |
|---|---|---|
| T1 | CI pipeline: validate → auto-tag → deploy, matching the sibling sites | done |
| T2 | Extend the leak check for a live-infrastructure source: account id, hostnames, stack FQDNs, EC2/AMI ids | done |
| T3 | Generate specs/ from machine-readable data, with a CI drift check | done |
| T4 | Cross-check every derived total against the survey's own summary blocks; fail the build on disagreement | done |
| T5 | The seven pages in the brief's build order, plus /security/ and /network/ | done |
| T6 | Publish the source documents verbatim, with the two redaction-list documents redacted under their own rule | done |
| T7 | llms-full.txt — the thing pki.sgit.ai lacks | done |
| T11 | The refocus, requested by the lead: lead the site with what has been built and shipped — the lifecycle, the generated CLIs, the AWS operator toolkit and its REPL/TUI, the sequence language, the UIs — so the site reads as an invitation to users and contributors. The audit stays published in full, one click away. Shipped in v0.2.0. | done |
| T8 | Per-page .md twins beyond index.md | queued |
| T9 | Wire the spec generator to the platform repository directly, so the catalogue is read from manifest.py rather than from a dated survey | queued |
| T10 | A Wardley map of the substrate spectrum, for wardley-maps.sgit.ai | queued |
Requests — these need an answer from a person
| # | Request | State |
|---|---|---|
| R1 | The brief pack's catalogue JSON is missing the per-spec array its own README advertises. specs__catalogue.json is described as “the full catalogue — LOC, capability, boot time, stability, family, create_node support” and ships with counts and totals but no specs array. The rows were transcribed from document 04 section 1 instead, and every derived total is cross-checked against the JSON's own summary blocks — they agree. Worth fixing at source so the next consumer does not have to transcribe a markdown table. | for the Librarian |
| R2 | The pack disagrees with itself on the stable/experimental split. Document 00 section 7 says “8 STABLE, 8 EXPERIMENTAL”; document 04 section 1 and the catalogue JSON both say 9 stable, 7 experimental, and the per-spec rows compute 9/7. The site publishes 9/7 and states its working. One line to correct. | for the Librarian |
| R3 | Is the platform still being developed? No commit since 24 July 2026. Q8 is published unresolved and the site deliberately makes no active-development claim, but this is the one open question the tree cannot answer. Paused, finished, or between phases? | for the lead |
| R4 | Confirm the domain decision. The site states, on the front page and in /network/, that sgit.ai is documentation and sgraph.ai is running infrastructure, and it has encoded that as the single exception in the leak tripwire. That is the pack's recommendation rather than a ratified decision — it should be ratified or reversed rather than left as a site's assumption. | for the lead |
| R5 | Q6, the EC2 tag. Published as a known trade-off with a named mitigation. The site says a stated decision either way is fine and silence is not — which puts the decision on this list. | for the lead |
| R6 | One docker images line closes G5 and lets the site publish an image size instead of an empty cell. | low priority |
Notes from the build
- The first release shipped without its own release gate. This repository was created from a Python
.gitignore, whosebuild/rule silently swallowedadmin/build/— every generator andvalidate.jspassed locally and were simply absent from the checkout, so v0.1.0's validate job died on a missing file before it could check anything. The sibling repositories carry a!admin/build/negation for exactly this; this one did not. Fixed in v0.1.1, and worth noting because it is the same failure mode the whole site is about: a check that appears to be running and is not. - The site's own rule caught something immediately. The derived-total cross-check in
gen_specs.pywas written expecting to be redundant; it is what confirmed the family LOC totals, the browser share and the 9/7 split all reconcile — and it is what makes R2 a documentation error rather than a data error. - Two source documents could not be published verbatim. Documents 07 and 10 are the redaction list, so between them they named every live address the site was told to strip. They are published redacted under their own rule, and the same list is now in the release gate.
- No word of this site was written from the platform's README. Per the pack's instruction, and because the README describes a repository that does not exist.