gpt-5 through gpt-5.3-codex (and the gpt-5.2-codex personality files) are
no longer served in models_cache.json; current lineup is gpt-5.5, gpt-5.4,
gpt-5.4-mini, gpt-5.3-codex-spark, codex-auto-review.
Replaces the lean file from #166 with the template verified byte-for-byte
against a fresh local models_cache.json (fetched 2026-07-08) and against
codex-rs/models-manager/models.json at openai/codex HEAD. Also matches the
SYSTEM INSTRUCTIONS section of gpt-5.5-full.md exactly.
- claude-design.md: expand from skill index to full bodies for all 13 listed
skills plus Make a Doc and Read PDF (+ starter component .jsx blocks)
- OpenAI/Codex/gpt-5.5.md: full prompt content
- README: note both updates in Recently Updated
- Normalize trailing whitespace (remove from code blocks, add outside)
- Wrap bare XML tags in backticks for GitHub rendering
- Convert invisible PUA content-reference delimiters to visible brackets
- Add paragraph spacing around structural XML tags
- Fix template placeholders (<variant>, <id>, <content>) to render on GitHub
Writing blocks: rewrite with MUST/NEVER rules, 3-block limit, fence
line rules, content reference restriction. Automations: replace
description with structured field definitions, 5 worked examples,
timing dispatch rules, hourly frequency cap. file_search: add explicit
source routing rules and connected-source restriction. Image policy:
add "say as much as you can instead of refusing" guidance.
From corrected truth ledger: add image/person content-policy block,
developer personality instruction section, and preserve raw typo
"pornagraphy" from live capture.
Adds timing_mode and display_description to automations, calendar_id
to gcal write ops, personal_context and api_tool namespaces, image_gen
deprecation comments and channel enforcement, job opportunities to
genui, source_filter to file_search, connected sources via api_tool,
renames learning_block_v3 to math_block_widget_always_prefetch_v2
(66→60 IDs), and updates developer prompt date.
Append two trailing spaces to each consecutive metadata line in the
header block of every codex file so the GitHub preview renders them
on separate lines instead of word-wrapping the whole block as one
paragraph. The last line before the blank-line/separator is left
alone since the blank line already breaks the paragraph.
Replaces old OpenAI/codex-cli.md and OpenAI/gpt-5.3-codex.md with a
structured OpenAI/codex/ folder: one file per model (12 models) plus
four personality files extracted from model_messages.instructions_variables
(friendly + pragmatic, with gpt-5.2-codex's slightly divergent variants
kept separate).
Added comprehensive instructions for handling various tasks, including document creation, trustworthiness guidelines, skill invocation rules, and persona guidelines.
Ásgeir Thor Johnson