agentsclimarketplace

Skill

Skill hunterx-code/accessible-codex-session/skill

Screen-reader-friendly, keyboard/voice-friendly Codex session behavior skill.

Install
npx -y skills add hunterx-code/accessible-codex-session --skill skill

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Accessibility-aware Codex session behavior for screen-reader, keyboard-only, voice-control, low-vision, reduced-motion, and cognitive-load-sensitive users. Use when the user asks for accessible Codex interaction, screen-reader-friendly output, disability accommodations, keyboard/voice-friendly workflow, reduced terminal noise, visual-content descriptions, accessible approval prompts, or help using Codex despite accessibility barriers. Also use when Codex work involves screenshots, visual diffs, dense command output, approval prompts, or workflows that may exclude assistive-technology users.

SKILL.md

6.7 KB, as published. Nobody here has run it

Accessible Codex Session

Purpose

Use this skill to adapt the agent's own session behavior so Codex work is easier to follow with screen readers, keyboard-only control, voice control, low vision, reduced-motion needs, or lower cognitive-load needs.

This is a session accommodation skill, not a WCAG audit suite, compliance checker, medical accommodation assessment, or native Codex UI patch. It cannot fix Codex focus bugs, Accessibility API exposure, terminal behavior, or screen- reader defects. It reduces avoidable friction in the parts the agent controls: structure, summaries, command-output handling, visual descriptions, and approval wording.

First Steps

  1. Identify the accommodation profile from the user's request if visible.
  2. If the profile is not visible and the task depends on it, ask one concise question. Do not make the user repeatedly restate their disability.
  3. State the working profile in one short line when it affects output.
  4. Continue the task normally, but apply the session rules below.

If the user asks for a summary, screenshot explanation, visual diff, approval rewrite, or log cleanup but has not provided the artifact, do not invent details. Provide the accessible response format and ask for the missing artifact or state that the answer is format-only.

Accommodation profiles:

  • screen-reader: blind or low-vision users using NVDA, JAWS, VoiceOver, Orca, or similar.
  • keyboard-voice: keyboard-only, switch access, dictation, or Talon/voice- control users.
  • low-vision: users who need clear text, named colors, zoom-safe steps, and no reliance on tiny visual differences.
  • low-noise: users affected by spinner noise, repeated announcements, excessive logs, motion, or terminal churn.
  • cognitive-load: users who need stable state, short chunks, explicit next actions, and reduced context switching.
  • mixed: apply the strictest useful subset when multiple needs are visible.

Session Rules

Structure Responses

  • Use short headings and plain bullets.
  • Prefer prose or lists over dense tables. Use a table only when it materially improves comparison, and keep columns few.
  • Put the main decision, current state, or next action near the top.
  • Keep routine accessible summaries under about 10 bullets or 250 words unless the user asks for detail or the risk requires more evidence.
  • Use stable labels such as Current State, Next Action, Approval Needed, Files Changed, and Verification.
  • Avoid decorative separators, box drawing, ASCII art, emoji-only status, or meaning carried only by color, indentation, or spatial layout.
  • Do not say "as shown above", "click the thing on the left", or "the red one" without a text equivalent.

Reduce Terminal Noise

  • Summarize long command output instead of pasting large raw logs.
  • Preserve exact errors, failing commands, file paths, test names, and exit status.
  • Strip ANSI escape codes, spinner frames, progress-only churn, and repeated identical lines when relaying output.
  • Use scripts/sr_text_filter.py when command output is noisy or when the user asks for screen-reader-friendly logs.
  • If the full log matters, save or point to the log file and provide a concise accessible summary first.

Describe Visual Artifacts

When working with screenshots, images, charts, UI diffs, visual layouts, or rendered pages:

  • Provide a text description of the relevant content before asking the user to decide.
  • Name layout, visible text, controls, state changes, color-dependent meaning, focus order, and any suspected overlap or clipping.
  • For diffs, describe what changed and why it matters; do not rely only on before/after images.
  • For charts, name axes, units, trends, outliers, uncertainty, and source notes.
  • If visual inspection is uncertain, say what remains uncertain and what would verify it.

Make Approvals Accessible

Before any external action, live system change, publish/upload, or irreversible operation, provide a short body-visible approval block:

Approval Needed
Action:
Target:
Account or identity:
Content or payload:
Attachments or links:
Not included:
Approval phrase:
Reject phrase:

Keep the approval surface short and exact. Do not hide critical details in a long paragraph. Stop on hesitation such as wait, not yet, stop, or do not send. Include a reject phrase whenever an approval phrase is included so keyboard, voice, and screen-reader users have an unambiguous way to stop the action.

Support Keyboard And Voice Workflows

  • Prefer commands, file paths, and numbered choices over pointer-based directions.
  • Avoid instructions that require dragging, precise mouse placement, or visual scanning unless no alternative exists.
  • Provide keyboard alternatives when using browser, app, or UI workflows.
  • For repeated steps, group commands into a small numbered sequence.
  • If a task requires visual review, offer a text-first summary and the smallest specific question that lets the user decide.

Lower Cognitive Load

  • Keep progress updates short and meaningful.
  • Say what changed since the last update instead of restating the whole plan.
  • Name blockers plainly and give the next concrete local action.
  • Avoid introducing new tools, files, or abstractions unless they reduce real complexity.
  • End with verification status and remaining risk when appropriate.

Routing

For web, app, document, PDF, or compliance accessibility audits, use a dedicated accessibility/WCAG/screen-reader testing skill when available. This skill may still be used alongside those tools to keep the Codex session itself accessible.

Do not claim this skill certifies accessibility or legal compliance. It is an agent behavior aid.

Bundled Resources

  • references/accommodation-patterns.md: detailed patterns and anti-patterns for each accommodation profile.
  • scripts/sr_text_filter.py: clean noisy terminal or log text for screen- reader-friendly review.

Use the reference file when the task involves multiple profiles, approval language, visual artifact review, or uncertainty about the right accommodation.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.