Downstream repos leak the private template repo, plus two prose/comment style-rule proposals

[ptr727]

Filed from a derived repo per the drift-reporting responsibility. Two independent items.

1. Drift: downstream files name and link the private template repo

Template-injected content in derived repos references the private template repo by name and link. That is a dead link for public readers, since they cannot open a private repo, and it leaks the coordinator's existence. Found in a derived repo:

• .github/copilot-instructions.md, the "When in Doubt" drift-reporting paragraph, named and linked the private repo and an AGENTS.md anchor in it.
• CODESTYLE.md, the spell-check-scope item, said "the template ships README + HISTORY as the default".

Suggest genericizing downstream-facing text so the private hub stays opaque. For example, route drift-reporting through "raise it with the maintainer" without naming or linking the private repo, and phrase defaults without attributing them to "the template". The agent-facing responsibility to report drift upstream can live in agent session memory rather than a public repo file.

I have scrubbed both locally in the derived repo, but the durable fix belongs here so it does not return on reconvergence.

2. Proposal: two general prose and comment style rules

These surfaced while the maintainer reviewed agent-authored text. Both are general, applying to any repo or language, so I am proposing them for the shared conventions rather than as a per-repo divergence.

• No class- or file-header summary comment blocks. A type gets a comment only for a specific non-obvious point, kept terse. This complements the existing "no design narrative" and one-line-comment rules.
• No semicolons as sentence separators in agent-authored prose (comments, docs, PR and commit text). Recast as a comma or as two sentences, the same house style as the existing em-dash ASCII ban. This is a prose rule only. Code statement terminators are unaffected.

Deferring on whether to adopt these, where they should live (AGENTS.md Comments and Character Set, or CODESTYLE.md), and how downstream repos apply them.

[ptr727]

Both items landed.

Item 1 (template references) - a genuinely good catch, in two halves.

The hard leak - carried files naming and linking the private template - was fixed in #359, which also mechanized a check in spec/audit.py so it cannot spread silently again. A fleet sweep found it in 13 of 20 repos, 8 of them public; the usual instance is exactly the drift-reporting paragraph you identified, whose second link targets an AGENTS.md section the hub no longer has, so it 404s even with access.

The soft half you found - prose attributing rules to "the template", which names no repo and so slips past a name-based check - was fixed in #360: eleven attributions across AGENTS.md, CODESTYLE.md, and WORKFLOW.md now state the rule directly ("README + HISTORY are the default", not "the template ships README + HISTORY as the default"). You were right that the durable fix belongs upstream or it returns on every reconvergence.

Two review passes caught attributions my first regex missed ("this template", "the release template"), which is the "fix the class, not the instance" rule earning its keep immediately.

Downstream delivery: folded into the open conformance issue where one existed (HomeAutomation-Config, KiCadLibrary, MediaTools) and filed a new issue in the other ten repos. PlexCleaner's is ptr727/PlexCleaner#867.

Item 2 (style rules) - both adopted, in the sections the owner chose.

• No class/type/file-header summary comment blocks -> AGENTS.md Comments, beside the existing design-narrative and verbosity-creep bans, since a header summary is the same failure at file scope. Added one carve-out: a license or provenance header a tool or policy requires is not a summary.
• No semicolon joining two independent clauses in agent prose -> AGENTS.md Character Set, beside the ASCII rules. Written to your stated scope ("as sentence separators"), so a semicolon separating items in a list that already contains commas keeps its standard use, and code terminators are untouched.

On sweeping: the hub carries 248 doc lines with prose semicolons, so a full rewrite of dense governance text was not worth the risk of changing meaning. The rule states that existing prose is corrected as each file is next edited rather than swept - honest about current state rather than claiming compliance the hub does not have.

Thanks - both items were well-evidenced, and item 1 in particular found a class the mechanical check could not.