Labyrinth exploration 1.1.0

The skill is universal: the same SKILL.md serves Codex and Claude Code.

• Native Codex support alongside Claude Code: interface metadata, discovery and invocation
  instructions, a client guide for subagents and permissions, and shared tutorial prompts.
• Dashboard handoffs can remain local or use an authorized private destination; client
  memory updates require the user's request, while project context stays in project files.
• Codex metadata validation and an opt-in native loader smoke test, with a reproducible
  procedure for isolated live research-workflow tests.
• Clarify that the public example's review metadata is illustrative and that a demo build
  does not supply its missing referee reports.
• Clarify the distinction between independently validating certificate code and replaying
  a new size, while retaining the T3 validation requirement and independent-review gate.
• assets/social-preview.png for GitHub's social preview, made by tools/screenshots.py social.
• The hero graphic and the social preview name no single client: "an agent skill" for Codex,
  Claude Code or any client that reads SKILL.md.

Labyrinth exploration 1.0.0

First public release.

• Method: the loop (read the frontier, explore, predict boldly, test and log, referee,
  update the map, look for shortcuts), the hard rules, the session protocol and the
  saturation ladder.
• Tiers and review states: T1–T6 keep proof apart from speculation; T2 and T3 results
  also carry a review state (unreviewed, under-review, refereed, human-checked).
• Campaigns: attack agents, literature agent, referees, writers and a coordinator;
  fill-in briefs in templates/briefs/.
• Engine (templates/lab.py, standard library):
  • a domain-neutral frontier map (frontier.json) with a sweep-line classifier and bins
    for large ranges;
  • detection of values claimed both realized and impossible;
  • the state-of-the-art table with history and escaped LaTeX and Markdown output;
  • validation of nodes, events, review states, frontier rows and table rows.
• Dashboard (templates/dashboard.html):
  • map, barcodes, state of the art, frontier, optional scatter, knowledge graph, timeline,
    board and tables;
  • light and dark mode;
  • every panel fails on its own, with a message;
  • ?tab= opens a tab directly.
• Example: triangle counts in graphs on n vertices, exact for n ≤ 7, with an exhaustive
  test for n = 8.
• Docs: README, step-by-step tutorial, example prompts, graphics generated by
  tools/make_graphics.py.

The method was developed on a research programme in pure mathematics in September and
October 2026; references/lessons.md records its lessons without its results.