Skip to content

fix(docs): review fixes for Authentication pilot (framework, nav, structure) - #15

Merged
rcha0s merged 2 commits into
mainfrom
docs/architectural-control-authentication
Sep 1, 2026
Merged

rcha0s merged 2 commits into
mainfrom
docs/architectural-control-authentication

Conversation

@rcha0s

@rcha0s rcha0s commented Sep 1, 2026

Copy link
Copy Markdown
Owner

Summary

Follow-up review fixes on top of the merged Authentication pilot (#14), on the same branch since #14 was merged before these landed:

Framework grounding (ADR-0003 + skill now require this for every future doc):

  • The mental-model blockquote skipped straight to "how platforms implement auth" without ever stating what authentication is. Rewritten to open with the three-factor framework (something you know / have / are, NIST SP 800-63-4) before anything else.
  • Removed a "diagram: omitted, " sentence that had leaked into the shipped doc as literal prose — that call is process metadata for the orchestrating conversation, not document content. Skill instructions updated to say so explicitly.
  • Removed a meta-commentary sentence opening "## Interviewer probes" that described the section's own format instead of starting on the first question.

Prose structure:

  • "Where this decision forks" and all four context sections had accumulated 3-4 back-to-back short paragraphs each making one parallel, independent point. Converted each run into a bulleted list with a bold lead-in per point; kept the worked compare-contrast examples (fintech vs. admin console, banking app vs. habit tracker, mesh vs. marketing platform) as prose since those read better as narrative.
  • The skill was missing the general CONCISENESS: principle that its sibling add-appsec-topic skill has as a named block (only had scattered doc-specific rules). Added it, explicit that it applies to prose only, never to table cells.

Nav bar:

  • The top tab bar now has 11 top-level sections and overflows its container at essentially every desktop width (confirmed: scrollWidth 1426 vs clientWidth 1216, consistent from ~1220px up) with a bare overflow-x scrollbar and no visual cue, so later tabs (Misconfiguration, Infrastructure, AI & LLM Security) were effectively undiscoverable. Added a CSS mask-image fade on both edges of .md-tabs__list, verified locally against a screenshot before shipping.

All of the skill/ADR changes here are locked in as requirements for every future architectural-control doc, not one-off fixes to this pilot.

Test plan

  • mkdocs build --strict — clean
  • pytest tests/ — 11/11 passing
  • Pilot doc citation consistency re-verified after both rounds of edits (cited == anchored, no gaps)
  • Nav fade fix screenshotted locally before shipping
  • Mid:/Principal: probe format confirmed intact (8/8 pairs), unchanged by this PR

🤖 Generated with Claude Code

…erflow

Three fixes from review of the Authentication pilot and the new
architectural-control doc type:

- The mental-model blockquote skipped straight to "how platforms
  implement auth" without ever stating what authentication actually
  is. Rewritten to open with the three-factor framework (something you
  know / have / are, NIST SP 800-63-4) before anything else. Locked
  this as a requirement for the doc type going forward, not just a
  one-off pilot fix: ADR-0003 and the add-architectural-control skill
  now require every doc's blockquote to name its topic's real
  first-principles framework (RBAC/ABAC/ReBAC for authorization, the
  CIA triad and Kerckhoffs's principle for encryption, etc.) before
  covering the fork and the reviewer's core check.
- A "diagram: omitted, <reason>" sentence, meant as process metadata
  reported back to the orchestrating conversation, had leaked into the
  shipped doc body as literal prose. Removed, and the skill's
  instructions now say explicitly that this call is response text, not
  document content.
- Long, TLDR-repeating paragraphs in "Where this decision forks" and
  the context sections tightened: cut one paragraph that only restated
  points the strengthened blockquote and adjacent paragraphs already
  made, split run-ons, fixed comma splices. The skill now requires
  short (2-4 sentence) paragraphs here and prohibits restating the
  blockquote in different words, matching the conciseness rule already
  in place for the other doc type.

Also: the top nav tab bar has 11 top-level sections now and overflows
its container at essentially every desktop width with a bare
overflow-x scrollbar and no visual cue, so the later tabs
(Misconfiguration, Infrastructure, AI & LLM Security) were effectively
undiscoverable. Added a CSS mask-image fade on both edges of
.md-tabs__list so an overflowing bar visibly signals there's more to
scroll to. Confirmed the overflow (scrollWidth 1426 vs clientWidth
1216, consistent from ~1220px up) and the fade rendering locally
before shipping.
…eness rule

Continues review of the Authentication pilot:

- The prose in "Where this decision forks" and each of the four
  context sections had accumulated 3-4 back-to-back short paragraphs
  each making one parallel, independent point. No single paragraph
  was long, but the run of them read as a wall. Converted each such
  run into a short bulleted list with a bold lead-in per point, kept
  every worked compare-contrast example (fintech vs admin console,
  banking app vs habit tracker, service mesh vs marketing platform)
  as prose, since those read better as flowing narrative than
  fragmented bullets.
- Removed a meta-commentary sentence opening "## Interviewer probes"
  that described the section's own format instead of starting on the
  first question.
- The skill was missing the general CONCISENESS principle that
  add-appsec-topic has as its own named block; it only had scattered
  doc-specific rules. Added the same principle here, explicit that it
  applies to prose only (blockquote and context-opening paragraphs),
  never to table cells, which are already dense by design.

Skill and post-run checklist updated to catch all of this on future
architectural-control docs: convert parallel-paragraph runs to
bulleted lists, no section-format meta-commentary, no prose paragraph
past 5-6 sentences.
@rcha0s
rcha0s merged commit 3716fc9 into main Sep 1, 2026
3 checks passed
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