Skip to content

refactor(skills): remove dated prompt patterns found by prompt audit #173

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ryzizub merged 3 commits into from
Oct 6, 2026
Merged

refactor(skills): remove dated prompt patterns found by prompt audit #173

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
232b582
refactor(skills): remove dated prompt patterns found by prompt audit
ryzizub Oct 5, 2026
a87342d
refactor(skills): resolve the audit's flag-only items
ryzizub Oct 5, 2026
1a60c47
docs: tighten the prose added by the audit
ryzizub Oct 5, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 4 additions & 4 deletions AGENTS.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -57,6 +57,7 @@ skills/ # every <skill>/ ships SKILL.md + agents/openai.yaml (C
platforms/web.md
platforms/windows.md
testing.md
wcag-criteria.md
widget-mapping.md
animations/SKILL.md
animations/references/
Expand Down Expand Up @@ -137,10 +138,9 @@ Every `SKILL.md` follows this structure:
- `argument-hint` _(optional)_ — placeholder hint shown to the user (e.g., `"[file-or-directory]"`)
- `model` _(optional)_ — model to use **while the skill is active**. The override applies for
the rest of the current turn and is not saved to settings, so a user on Opus who triggers a
skill pinned to `sonnet` stays on sonnet until their next prompt. Nine skills here set it.
Note the eval consequence: `create-project` pins `haiku`, so in a two-arm run its with-plugin
arm answers on a weaker model than the no-plugin arm, which understates its measured lift.
Routing is decided before the switch, so a pin never explains a routing miss
skill pinned to `sonnet` stays on sonnet until their next prompt. No skill here sets it.
A pin weaker than the session model lowers that skill's eval Δ; it never explains a
routing miss, since routing is decided before the switch
- `effort` _(optional)_ — reasoning effort while the skill is active (`low`/`medium`/`high`/
`xhigh`/`max`), overriding the session level for that turn. Seven skills here set it
2. **H1 title** — human-readable skill name
Expand Down
9 changes: 1 addition & 8 deletions agents/flutter-reviewer.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,8 +72,7 @@ report"). If it is unavailable, rely on the four preloaded standards alone.
### When scoping fails

If you cannot determine a change scope — not a git repository, detached HEAD, no merge base, or the
git commands fail — report that you could not determine a change scope and stop. Do not guess and do
not review the whole repository.
git commands fail — report that you could not determine a change scope and stop. Do not guess.

## Output

Expand Down Expand Up @@ -127,9 +126,3 @@ coverage. For example:
separately by the plugin's PostToolUse `analyze.sh` hook when code is written, not here. Use the
analyzer only to corroborate a skill-based judgment.
- **Untraceable findings.** If a finding cannot name one of the four loaded standards, omit it.

## Dispatch contract

When dispatched by an orchestrator or critic round, you self-scope via the adaptive diff procedure
above — the caller does not pass you a file list — and the caller consumes your findings table
verbatim.
6 changes: 3 additions & 3 deletions evals/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -377,9 +377,9 @@ Read the two arm scores, not the total. The without-arm is supposed to score bad
- **The no-plugin arm swings between runs.** One two-arm run cannot disqualify a grader:
`accessibility-declines-gesture-detector-tap-target` failed all three content graders
without the plugin on one run and passed all three on the next.
- **`create-project` pins `model: haiku`.** Its with-arm answers on a weaker model than its
baseline, so its Δ reads low. Routing is decided before the switch, so the pin never
explains a routing miss.
- **A `model` pin weaker than the session model lowers that skill's Δ.** The with-arm
answers on the pinned model, the baseline on the session model. It never explains a
routing miss.

Costs: **$13** for 102 cases in one arm, **$25** for both, roughly **$0.12 per run**. At
`--runs 3` a two-arm sweep is six runs per case, so budget around **$75**. `-j` up to 8
Expand Down
77 changes: 4 additions & 73 deletions skills/accessibility/SKILL.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,7 +19,7 @@ Flutter accessibility auditing and remediation across WCAG 2.2 conformance level
AAA. This file is the workflow; the reference files listed under **Additional Resources** are
loaded on demand, and each phase below names the one it needs.

Load only the file the current phase names. Do not duplicate its content here.
Load only the file the current phase names.

---

Expand Down Expand Up @@ -51,7 +51,7 @@ Apply these standards to all accessibility work:

**Text Containers** (WCAG 1.4.4) — Fixed-height containers must not wrap `Text`. Use `minHeight` constraints. Fixed heights clip text at 1.5x font scale on Android, sooner on iOS where Larger Accessibility Sizes go to ~3.1x.

**Contrast** (WCAG 1.4.3) — All text and UI components must meet the contrast ratio for the selected WCAG level. See the WCAG Level Criteria Reference section below.
**Contrast** (WCAG 1.4.3) — All text and UI components must meet the contrast ratio for the selected WCAG level. See [`references/wcag-criteria.md`](references/wcag-criteria.md).

**Cupertino Semantics** (WCAG 4.1.2) — Cupertino widgets (`CupertinoSwitch`, `CupertinoSlider`, `CupertinoSegmentedControl`, `CupertinoButton`) ship with weaker semantic defaults than their Material equivalents. Always wrap them in `Semantics(label:, value:, button:)`.

Expand Down Expand Up @@ -203,77 +203,7 @@ Write tests covering every fix so it cannot regress. See [`references/testing.md

## WCAG 2.2 Level Criteria Reference

Level AA includes all Level A criteria. Level AAA includes all Level A and AA criteria. The "Version" column flags whether the criterion is from WCAG 2.0, 2.1, or 2.2. WCAG 2.2 removed 4.1.1 Parsing.

### Level A

| WCAG ID | Version | Criterion | Flutter Check |
| --- | --- | --- | --- |
| 1.1.1 | 2.0 | Non-text Content | `semanticLabel` on images, `Semantics(label:)` on icons, `excludeFromSemantics: true` on decorative |
| 1.3.1 | 2.0 | Info and Relationships | Semantic roles. `MergeSemantics` for grouped label/value pairs only, never around interactive children |
| 1.3.2 | 2.0 | Meaningful Sequence | Reading order matches visual order. `FocusTraversalGroup` + `OrderedTraversalPolicy` |
| 1.3.3 | 2.0 | Sensory Characteristics | Instructions do not rely solely on shape, size, location, or sound |
| 1.4.1 | 2.0 | Use of Color | Color never sole differentiator |
| 2.1.1 | 2.0 | Keyboard | All functionality via keyboard or switch access. No bare `GestureDetector` |
| 2.1.2 | 2.0 | No Keyboard Trap | Focus can always be moved away |
| 2.3.1 | 2.0 | Three Flashes or Below Threshold | No content flashes > 3 times per second |
| 2.4.1 | 2.0 | Bypass Blocks | Skip-navigation mechanism. **Web only** |
| 2.4.2 | 2.0 | Page Titled | Each screen has a meaningful title. **Web: `<title>` tag** |
| 2.4.3 | 2.0 | Focus Order | Tab/focus order preserves meaning |
| 2.5.3 | 2.1 | Label in Name | Visible label text contained in accessible name |
| 3.2.6 | 2.2 | Consistent Help | Help mechanism in same relative order across screens |
| 3.3.1 | 2.0 | Error Identification | Form errors identified in text, not color alone |
| 3.3.2 | 2.0 | Labels or Instructions | All form fields have visible labels |
| 3.3.7 | 2.2 | Redundant Entry | Multi-step forms must not re-collect already-provided info unless re-entry is essential |
| 4.1.2 | 2.0 | Name, Role, Value | `Semantics(label:, button: true)`, `Tooltip`, state via `checked`, `selected`, `enabled` |
| 4.1.3 | 2.1 | Status Messages | `Semantics(liveRegion: true)`, `SemanticsService.announce()` |

### Level AA (adds these to Level A)

| WCAG ID | Version | Criterion | Flutter Check |
| --- | --- | --- | --- |
| 1.3.4 | 2.1 | Orientation | App not locked to single orientation without essential reason |
| 1.3.5 | 2.1 | Identify Input Purpose | Correct `keyboardType` and `autofillHints` |
| 1.4.3 | 2.0 | Contrast (Minimum) | Normal text 4.5:1, large text 3:1 |
| 1.4.4 | 2.0 | Resize Text | Text scales to 200% (300% on iOS) without loss |
| 1.4.5 | 2.0 | Images of Text | Use `Text`, not images of text |
| 1.4.10 | 2.1 | Reflow | Content reflows at 320 CSS px equivalent |
| 1.4.11 | 2.1 | Non-text Contrast | UI components and focus indicators 3:1 |
| 1.4.12 | 2.1 | Text Spacing | Content not lost under increased spacing |
| 1.4.13 | 2.1 | Content on Hover or Focus | Dismissable, hoverable, persistent. **Web/desktop** |
| 2.4.5 | 2.0 | Multiple Ways | More than one way to locate a screen |
| 2.4.6 | 2.0 | Headings and Labels | Descriptive. `Semantics(header: true)` for sections |
| 2.4.7 | 2.0 | Focus Visible | Keyboard focus indicator always visible |
| 2.4.11 | 2.2 | Focus Not Obscured (Minimum) | Focused component not entirely hidden by author-created content |
| 2.5.7 | 2.2 | Dragging Movements | Every drag has a single-pointer alternative |
| 2.5.8 | 2.2 | Target Size (Minimum) | Targets >= 24x24 CSS px, with documented exceptions |
| 3.1.2 | 2.0 | Language of Parts | **Web: `lang` attribute** |
| 3.2.3 | 2.0 | Consistent Navigation | Consistent across screens |
| 3.2.4 | 2.0 | Consistent Identification | Same-function components identified consistently |
| 3.3.3 | 2.0 | Error Suggestion | Suggested correction when possible |
| 3.3.4 | 2.0 | Error Prevention (Legal/Financial) | Reversible or confirmable |
| 3.3.8 | 2.2 | Accessible Authentication (Minimum) | No required cognitive function test without alternative. Allow paste, support password managers, no puzzle CAPTCHAs without alternative |

### Level AAA (adds these to A and AA)

| WCAG ID | Version | Criterion | Flutter Check |
| --- | --- | --- | --- |
| 1.4.6 | 2.0 | Contrast (Enhanced) | Normal 7:1, large 4.5:1 |
| 2.1.3 | 2.0 | Keyboard (No Exception) | No `GestureDetector` anywhere |
| 2.2.3 | 2.0 | No Timing | No time limits except real-time events |
| 2.2.6 | 2.1 | Timeouts | Inactivity warning |
| 2.3.2 | 2.0 | Three Flashes | Zero flashing |
| 2.3.3 | 2.1 | Animation from Interactions | Every animation gated on `disableAnimations` |
| 2.4.8 | 2.0 | Location | Users always know where they are |
| 2.4.9 | 2.0 | Link Purpose (Link Only) | Understandable from link text alone |
| 2.4.12 | 2.2 | Focus Not Obscured (Enhanced) | No occlusion at all, not just total occlusion |
| 2.4.13 | 2.2 | Focus Appearance | At least 2 CSS px perimeter, encloses component, 3:1 against unfocused |
| 2.5.5 | 2.1 | Target Size (Enhanced) | Targets >= 44x44 CSS px |
| 2.5.6 | 2.1 | Concurrent Input Mechanisms | No single-modality restriction |
| 3.2.5 | 2.0 | Change on Request | Context changes only on user request |
| 3.3.5 | 2.0 | Help | Context-sensitive help available |
| 3.3.6 | 2.0 | Error Prevention (All) | All submissions reversible or confirmable |
| 3.3.9 | 2.2 | Accessible Authentication (Enhanced) | No cognitive function test even with alternative |
Per-level criterion tables with the Flutter check for each: [`references/wcag-criteria.md`](references/wcag-criteria.md). Load it in Phase 3.

---

Expand Down Expand Up @@ -349,6 +279,7 @@ For corrected snippets, full classes (`AccessibleTapTarget`, `AccessibleSlider`,

## Additional Resources

- [`references/wcag-criteria.md`](references/wcag-criteria.md) — per-level WCAG 2.2 criterion tables with the Flutter check for each.
- [`references/audit-templates.md`](references/audit-templates.md) — severity guide, report template, level-specific passed-check lists, cross-platform severity table.
- [`references/examples.md`](references/examples.md) — full Flutter widget classes per category, including all WCAG 2.2 patterns.
- [`references/widget-mapping.md`](references/widget-mapping.md) — widget-to-requirement quick reference for all commonly audited Flutter widgets.
Expand Down
75 changes: 75 additions & 0 deletions skills/accessibility/references/wcag-criteria.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,75 @@
# WCAG 2.2 Level Criteria Reference

Every criterion by level, with the Flutter check that satisfies it.

Level AA includes all Level A criteria. Level AAA includes all Level A and AA criteria. The "Version" column flags whether the criterion is from WCAG 2.0, 2.1, or 2.2. WCAG 2.2 removed 4.1.1 Parsing.

## Level A

| WCAG ID | Version | Criterion | Flutter Check |
| --- | --- | --- | --- |
| 1.1.1 | 2.0 | Non-text Content | `semanticLabel` on images, `Semantics(label:)` on icons, `excludeFromSemantics: true` on decorative |
| 1.3.1 | 2.0 | Info and Relationships | Semantic roles. `MergeSemantics` for grouped label/value pairs only, never around interactive children |
| 1.3.2 | 2.0 | Meaningful Sequence | Reading order matches visual order. `FocusTraversalGroup` + `OrderedTraversalPolicy` |
| 1.3.3 | 2.0 | Sensory Characteristics | Instructions do not rely solely on shape, size, location, or sound |
| 1.4.1 | 2.0 | Use of Color | Color never sole differentiator |
| 2.1.1 | 2.0 | Keyboard | All functionality via keyboard or switch access. No bare `GestureDetector` |
| 2.1.2 | 2.0 | No Keyboard Trap | Focus can always be moved away |
| 2.3.1 | 2.0 | Three Flashes or Below Threshold | No content flashes > 3 times per second |
| 2.4.1 | 2.0 | Bypass Blocks | Skip-navigation mechanism. **Web only** |
| 2.4.2 | 2.0 | Page Titled | Each screen has a meaningful title. **Web: `<title>` tag** |
| 2.4.3 | 2.0 | Focus Order | Tab/focus order preserves meaning |
| 2.5.3 | 2.1 | Label in Name | Visible label text contained in accessible name |
| 3.2.6 | 2.2 | Consistent Help | Help mechanism in same relative order across screens |
| 3.3.1 | 2.0 | Error Identification | Form errors identified in text, not color alone |
| 3.3.2 | 2.0 | Labels or Instructions | All form fields have visible labels |
| 3.3.7 | 2.2 | Redundant Entry | Multi-step forms must not re-collect already-provided info unless re-entry is essential |
| 4.1.2 | 2.0 | Name, Role, Value | `Semantics(label:, button: true)`, `Tooltip`, state via `checked`, `selected`, `enabled` |
| 4.1.3 | 2.1 | Status Messages | `Semantics(liveRegion: true)`, `SemanticsService.announce()` |

## Level AA (adds these to Level A)

| WCAG ID | Version | Criterion | Flutter Check |
| --- | --- | --- | --- |
| 1.3.4 | 2.1 | Orientation | App not locked to single orientation without essential reason |
| 1.3.5 | 2.1 | Identify Input Purpose | Correct `keyboardType` and `autofillHints` |
| 1.4.3 | 2.0 | Contrast (Minimum) | Normal text 4.5:1, large text 3:1 |
| 1.4.4 | 2.0 | Resize Text | Text scales to 200% (300% on iOS) without loss |
| 1.4.5 | 2.0 | Images of Text | Use `Text`, not images of text |
| 1.4.10 | 2.1 | Reflow | Content reflows at 320 CSS px equivalent |
| 1.4.11 | 2.1 | Non-text Contrast | UI components and focus indicators 3:1 |
| 1.4.12 | 2.1 | Text Spacing | Content not lost under increased spacing |
| 1.4.13 | 2.1 | Content on Hover or Focus | Dismissable, hoverable, persistent. **Web/desktop** |
| 2.4.5 | 2.0 | Multiple Ways | More than one way to locate a screen |
| 2.4.6 | 2.0 | Headings and Labels | Descriptive. `Semantics(header: true)` for sections |
| 2.4.7 | 2.0 | Focus Visible | Keyboard focus indicator always visible |
| 2.4.11 | 2.2 | Focus Not Obscured (Minimum) | Focused component not entirely hidden by author-created content |
| 2.5.7 | 2.2 | Dragging Movements | Every drag has a single-pointer alternative |
| 2.5.8 | 2.2 | Target Size (Minimum) | Targets >= 24x24 CSS px, with documented exceptions |
| 3.1.2 | 2.0 | Language of Parts | **Web: `lang` attribute** |
| 3.2.3 | 2.0 | Consistent Navigation | Consistent across screens |
| 3.2.4 | 2.0 | Consistent Identification | Same-function components identified consistently |
| 3.3.3 | 2.0 | Error Suggestion | Suggested correction when possible |
| 3.3.4 | 2.0 | Error Prevention (Legal/Financial) | Reversible or confirmable |
| 3.3.8 | 2.2 | Accessible Authentication (Minimum) | No required cognitive function test without alternative. Allow paste, support password managers, no puzzle CAPTCHAs without alternative |

## Level AAA (adds these to A and AA)

| WCAG ID | Version | Criterion | Flutter Check |
| --- | --- | --- | --- |
| 1.4.6 | 2.0 | Contrast (Enhanced) | Normal 7:1, large 4.5:1 |
| 2.1.3 | 2.0 | Keyboard (No Exception) | No `GestureDetector` anywhere |
| 2.2.3 | 2.0 | No Timing | No time limits except real-time events |
| 2.2.6 | 2.1 | Timeouts | Inactivity warning |
| 2.3.2 | 2.0 | Three Flashes | Zero flashing |
| 2.3.3 | 2.1 | Animation from Interactions | Every animation gated on `disableAnimations` |
| 2.4.8 | 2.0 | Location | Users always know where they are |
| 2.4.9 | 2.0 | Link Purpose (Link Only) | Understandable from link text alone |
| 2.4.12 | 2.2 | Focus Not Obscured (Enhanced) | No occlusion at all, not just total occlusion |
| 2.4.13 | 2.2 | Focus Appearance | At least 2 CSS px perimeter, encloses component, 3:1 against unfocused |
| 2.5.5 | 2.1 | Target Size (Enhanced) | Targets >= 44x44 CSS px |
| 2.5.6 | 2.1 | Concurrent Input Mechanisms | No single-modality restriction |
| 3.2.5 | 2.0 | Change on Request | Context changes only on user request |
| 3.3.5 | 2.0 | Help | Context-sensitive help available |
| 3.3.6 | 2.0 | Error Prevention (All) | All submissions reversible or confirmable |
| 3.3.9 | 2.2 | Accessible Authentication (Enhanced) | No cognitive function test even with alternative |
Loading
Loading