feat(han-communication): honor the shape the reader asked for

CONTRIBUTING.md

@@ -255,7 +255,7 @@ test, add it to that list and wire the standard in:
    skill's real reader is a specific expert (an engineer, a pull-request reviewer, a non-technical stakeholder), name
    that reader instead of defaulting. Scope the frame per section so technical specifics the reader needs are not
    simplified away.
-4. **Add the standardized self-check.** Before presenting, the skill runs six behaviorally-anchored yes/no criteria over
+4. **Add the standardized self-check.** Before presenting, the skill runs behaviorally-anchored yes/no criteria over
    the prose regions only: main point first, descriptive headings, one idea per paragraph, sentence length, common
    words with no blocklisted word and an explanation for every term the reader cannot look up, every fact preserved. It
    corrects any failure. Leave code fences, diagram bodies, rendered markup, and citation identifiers unevaluated and

docs/concepts.md

@@ -188,10 +188,10 @@ writes. That rule makes the deliverable lead with its main point, give each para
 headings, keep sentences short and active, prefer common words, and reveal detail in layers.
 
 The rule is applied in stages, never as one instruction block. Its structural rules shape each skill's output template,
-and its six behaviorally-anchored criteria run as a discrete self-check after the draft exists. Skills with a synthesis
-or editor step also dispatch the [`readability-editor`](../han-communication/docs/agents/readability-editor.md) agent to
+and its behaviorally-anchored criteria run as a discrete self-check after the draft exists. Skills with a synthesis or
+editor step also dispatch the [`readability-editor`](../han-communication/docs/agents/readability-editor.md) agent to
 rewrite the draft, preserving every fact. Fidelity outranks readability: no required fact is dropped to read more
-simply.
+simply, unless the reader asked for less and losing it would not change what they do next.
 
 Readability applies to the reader-facing skills (`/research`, `/gap-analysis`, `/project-documentation`,
 `/issue-triage`, `/runbook`, `/architectural-decision-record`, `/code-overview`, `/investigate`, `/code-review`,

docs/plans/readability-reader-format-requests/artifacts/.discovery-notes.md

@@ -0,0 +1,208 @@
+# Discovery notes
+
+Gathered 2026-08-19, before the interview. Paths are repo-relative.
+
+## The two files the issue names
+
+- `han-communication/references/readability-rule.md` (153 lines). Carries the audience frame, nine output
+  properties, length guidance, the prose-only scope, `## Fidelity wins` (lines 97-102), the scoped escape
+  clause (104-113), and `## The standardized self-check` (115-132). Line 130 reads "The set is enumerated,
+  not illustrative: these six criteria are the whole check."
+- `han-communication/output-styles/han-readability.md` (97 lines). A distilled copy. Its own
+  `## Fidelity wins` (71-75) and its own six-criterion check (85-97). The closure sentence "These six criteria
+  are the whole check" opens that block at line 87 rather than closing it.
+
+## The count is echoed on 25 more surfaces
+
+> **Superseded.** The counts in this section are what the pre-interview sweep found. The corrected inventory
+> is the touch-points table under "Implementation discovery" below, re-verified with wrap-tolerant searches at
+> synthesis on 2026-08-19. This section is kept because the specification cites it.
+
+The phrase "six-point self-check", "six-point checklist", or "six criteria" appears in **21 SKILL.md files**:
+
+```
+han-coding/skills/architectural-analysis      han-planning/skills/iterative-plan-review
+han-coding/skills/automated-test-planning     han-planning/skills/plan-a-feature
+han-coding/skills/code-overview               han-planning/skills/plan-a-phased-build
+han-coding/skills/coding-standard             han-planning/skills/plan-implementation
+han-coding/skills/design-an-api               han-planning/skills/plan-work-items
+han-coding/skills/investigate                 han-reporting/skills/html-summary
+han-coding/skills/manual-test-planning        han-reporting/skills/stakeholder-summary
+han-documentation/skills/architectural-decision-record   han-research/skills/gap-analysis
+han-documentation/skills/project-documentation           han-research/skills/issue-triage
+han-documentation/skills/runbook                         han-research/skills/research
+han-github/skills/update-pr-description
+```
+
+Several of those also name "criterion 6" as the fidelity guard by number, which the seventh criterion
+does not move but which is the same class of hardcoded reference.
+
+And in **3 operator-facing docs**: `docs/readability.md:105`,
+`han-communication/docs/output-styles/han-readability.md:72`,
+`han-communication/docs/agents/readability-editor.md:21,72`. The third one describes the editor's own rubric,
+which this change leaves alone, so only the first two are in the sweep (see F6).
+
+The canonical `han-communication/references/explanation-rule.md:17` also names the size, reading "a six-item
+self-check over a whole document". It sat outside this inventory and was added during review (F6).
+
+`han-communication/skills/readability-guidance/SKILL.md` is already count-free. It says "the standardized
+self-check" and "the fidelity criterion", never a number. It is the model for what the other 21 could be.
+
+## The editor agent runs a different six
+
+`han-communication/agents/readability-editor.md:95` says "Audit and rewrite against these six criteria.
+They are the whole rubric." Its set is **not** the rule's set:
+
+| # | Rule's self-check | Editor's rubric |
+| - | ----------------- | --------------- |
+| 1 | Main point first | Main point first |
+| 2 | Descriptive headings | Descriptive headings |
+| 3 | One idea per paragraph | One idea per paragraph |
+| 4 | Sentence length | Short, active sentences |
+| 5 | Common words, no blocklisted word | Common words, no blocklisted words |
+| 6 | Every fact preserved | Progressive disclosure |
+
+The editor carries fact preservation as an absolute principle above the rubric (lines 35-39) plus a
+fact-preservation ledger in its report, rather than as a rubric row.
+
+The editor also never receives a reader's shape request. Every dispatching skill passes it a file path
+and a named audience, and nothing else. `han-planning/skills/plan-a-feature/SKILL.md:432` is
+representative: "Pass the editor the file path ... and the named audience".
+
+## A prior decision already rejected a seventh criterion
+
+`docs/plans/orwell-six-rules/artifacts/implementation-decision-log.md:78-85` records decision D-4. It
+added two new principles to the editor agent beside the fidelity principle rather than as a seventh
+rubric criterion, and names the rejected alternative in its own words:
+
+> Add a seventh rubric criterion, rejected because it breaks the keep-it-small design (research O1) and
+> falsifies the "six criteria" count on the seven-plus surfaces that echo it (C6, F10).
+
+That plan measured "seven-plus surfaces". The grep above finds 25 outside the two canonical files. The
+prior decision is evidence about the cost, not a veto: issue #177 reopens it deliberately.
+
+## The rule forbids subjective criteria
+
+`readability-rule.md:117-118`: the self-check "evaluates concrete, behaviorally-anchored yes/no criteria,
+never 'is this clear?'" This bears directly on the issue's third proposal, a simplicity test beside the
+sentence-length ceiling.
+
+## The keep-it-small design principle is stated, not incidental
+
+`readability-rule.md:130-131`: the set "is kept small on purpose so it applies as one focused pass rather
+than decaying under its own weight." Line 70-72 states the same principle for the output properties.
+
+---
+
+# Implementation discovery
+
+Gathered 2026-08-19 by `plan-implementation`, appended rather than overwriting: the specification cites the
+inventory above, so replacing this file would break that reference.
+
+## Tech stack and tooling
+
+Markdown and Bash only. No application build, no dev server. The root `package.json` carries dev tooling.
+
+- **Lint:** `npm run lint` runs `prek run --all-files` — Prettier, ShellCheck, and file-hygiene hooks.
+- **Test:** `npm test` runs Bats over every `*.bats` file outside `node_modules`.
+- **Prettier settings:** `printWidth: 120`, `proseWrap: preserve`, `embeddedLanguageFormatting: off`. Prose
+  wrapping is preserved, so hand-wrapped paragraphs stay as written and reflowing is the author's job. **This is
+  why every inventory search here joins lines before matching**: a sentence routinely spans two lines, and a
+  line-oriented search silently misses it (D-10).
+
+## No automated test covers what this change edits
+
+Seven Bats files exist and every one covers a shell script:
+
+```
+test/sanity.bats                                          (harness proof only)
+scripts/han-config-dir.bats
+.claude/skills/han-release/scripts/remote-tag-state.bats
+han-coding/skills/code-review/scripts/detect-review-context.bats
+han-coding/skills/automated-test-planning/scripts/detect-test-context.bats
+han-planning/skills/iterative-plan-review/scripts/check-cross-references.bats
+han-planning/skills/plan-a-feature/scripts/verify-design-images.bats
+```
+
+Nothing asserts anything about the readability rule's text, the output style's text, or the skill files that
+quote them. The only automated guard on this change is Prettier formatting. **Verifying that a 28-file sweep
+landed completely is therefore an open implementation question, not a solved one.**
+
+## Touch points
+
+**The two canonical files, four passages each.**
+
+- `han-communication/references/readability-rule.md` (153 lines): `## Fidelity wins` (97-102), the escape
+  clause (104-113, whose closing sentence names criterion 5 positionally at line 112), `## The standardized
+  self-check` (115-132, whose closing paragraph declares the set closed at 130 and names criterion 6
+  positionally at 132).
+- `han-communication/output-styles/han-readability.md` (97 lines): the fidelity sentence at line 11,
+  `## Fidelity wins` (71-75), the escape clause (77-83, whose limit is worded differently from the rule's:
+  "It never licenses a blocked word and never licenses a lost fact"), and the check (85-97, declaring the set
+  closed at 87).
+
+**The quoting surfaces, verified by repository-wide search.**
+
+**Re-verified at synthesis on 2026-08-19 with wrap-tolerant searches.** The table below is the corrected
+inventory; where round 1's figure differed, the row says so.
+
+| Class | Count | Where |
+| ----- | ----- | ----- |
+| Size reference to the standard's self-check | 21 `SKILL.md` files, 25 occurrences | Six plugins; full list in the inventory above. Four files carry two occurrences each |
+| Size reference outside the skill directories | 5 files, 6 occurrences | `docs/readability.md:89` and `:105`, `docs/concepts.md:191`, `CONTRIBUTING.md:258`, `han-communication/docs/output-styles/han-readability.md:72`, `han-communication/references/explanation-rule.md:17`. **Corrected at synthesis:** round 1 recorded 3 files. The first three above were missed because the search matched only the hyphenated forms of the phrase. See D-11 |
+| Fidelity restatement naming **the standard** (changes) | 25 occurrences across 23 files | 20 self-check restatements in 20 `SKILL.md` files; 3 audience-frame paragraphs that still say "the standard governs" (`code-review`, `code-overview:59`, `stakeholder-summary`); `han-coding/skills/code-review/references/output-verification.md:97`; `docs/readability.md:155`. **Corrected at synthesis:** round 1 recorded 18. See D-4 |
+| Fidelity restatement naming **the frame** (unchanged) | 9 occurrences across 9 `SKILL.md` files | `coding-standard`, `automated-test-planning`, `readability-guidance`, `architectural-decision-record`, `iterative-plan-review`, `plan-a-feature`, `plan-a-phased-build`, `plan-implementation`, `plan-work-items`, plus the canonical sentence in `readability-rule.md`. **Corrected at synthesis:** round 1 recorded 8. See D-4 |
+| Positional reference to criterion 6 | 6 `SKILL.md` files + `readability-rule.md:132` | architectural-decision-record, runbook, issue-triage, html-summary, plan-work-items, iterative-plan-review |
+| Positional reference to criterion 5 (unchanged) | 1 `SKILL.md` file + `readability-rule.md:112` | `han-coding/skills/code-overview/SKILL.md:397`. Criterion 5 keeps its position and its meaning, so nothing about it became untrue. See D-12 |
+| Hardcoded enumeration of the whole check | 1 file | `han-coding/skills/code-review/references/output-verification.md:87-95` lists criteria 1 through 6 in its own words. **Found at synthesis;** absent from round 1's inventory. See D-13 |
+
+Totals: **63 corrections across 28 quoting files**, plus 8 passages in the 2 canonical files. Of the 21 skill
+files carrying a size reference, 20 also need the self-check fidelity fix and 1 (`plan-a-feature`) needs the
+size fix alone.
+
+**The fidelity class splits by grammatical subject, and only one side changes.** Twenty-five sites name *the
+standard* as what never drops a fact, and those change, because a reader's stated request can now drop one.
+Nine sites name *the frame*, and those stay: the frame still never decides whether a fact appears. Round 1
+recorded the split against the block a sentence sits in, which misroutes four sentences that sit in
+audience-frame paragraphs and still name the standard. `readability-guidance` and `plan-a-feature` carry only
+the frame form and take no fidelity edit at all. Recorded as D-4, corrected at synthesis.
+
+**Left alone deliberately.** Three files describe the readability editor's own rubric, which this change does
+not touch, so every statement in them stays true: `han-communication/agents/readability-editor.md` (criterion
+5 named positionally at lines 26, 27, 44, 142, and the rubric's size at 95),
+`han-communication/docs/agents/readability-editor.md` (three size references), and
+`han-communication/skills/edit-for-readability/SKILL.md` ("Do not restate the six rubric criteria here").
+**The third was added at synthesis.** A completeness search must also tolerate matches that have nothing to do
+with the readability check at all: `coding-standard`'s six adoption-bias checks, `edge-case-explorer`'s six
+dimensions, `gap-analyzer`'s six steps, and the "Six places" sentence in every vendored copy of
+`collaborative-stop-rule.md`. See D-1.
+
+## Precedent for this exact shape of change
+
+`docs/plans/orwell-six-rules/` planned the closest prior change: edits to the same standard and the same editor
+agent, with the same worry about count claims going stale. Two things it decided are worth carrying:
+
+- Its D-4 rejected a seventh rubric criterion **because** it would falsify the count on the surfaces that echo
+  it. This plan does the opposite, having removed that cost by going count-free instead.
+- Its D-7 ran a scoped documentation-consistency check limited to statements the edits made false, using the
+  repository's own `han-update-documentation` skill on the branch. That skill exists, scopes itself to what the
+  branch touched, and is a candidate verification step here.
+
+## Recent churn
+
+The readability area is among the most-edited in the plugin over 90 days:
+`readability-guidance/SKILL.md` 12 commits, `edit-for-readability/SKILL.md` 10,
+`docs/agents/readability-editor.md` 7, `agents/readability-editor.md` 6,
+`references/readability-rule.md` 5, `references/writing-voice.md` 4. Edits here meet actively-moving text.
+
+## Gaps found
+
+- **No coding standards document** for prose or reference files beyond `CLAUDE.md`'s Conventions section and
+  the `han-plugin-builder` authoring guidance.
+- **One ADR only:** `docs/adr/0001-project-configurable-default-swarm-size.md`. Nothing records the readability
+  standard's keep-it-small design closure as an architectural decision, even though this change reopens it.
+- **No test coverage** for the edited text, as recorded above.
+- **No versioning decision recorded.** The change touches seven plugins' shipped files. **Corrected in round 1:**
+  this section first said nine; the verified inventory lands in `han-coding`, `han-planning`, `han-documentation`,
+  `han-research`, `han-reporting`, `han-github`, and `han-communication`, and `docs/readability.md` is repo-root
+  rather than a plugin. Settled as D-6: no version moves on this branch.