Skip to content

docs: one PyAutoLabs AI-options page for every assistant; simplify READMEs, retire legacy #87

Closed

Description

@Jammy2211

Overview

Every PyAuto science assistant (autolens, autogalaxy, autofit, autocti) carries its own copy of the "which AI do I use" story, autogalaxy and autocti link into autolens_assistant's setup pages, and the free-option text (OpenCode "experimental", Gemini CLI stubs) is out of date: Gemini CLI's free tier ended 2026-06-18. This task creates one PyAutoLabs-level page in PyAutoScientist (mirroring the existing AI_POLICY.md stub pattern), shortens every assistant README to a link plus the clone snippet and starting prompt, and retires the chat/OpenCode/Gemini legacy. Decided by the human on 2026-10-23: three agentic routes (Claude Code, Codex, free = GitHub Copilot CLI and OpenCode, both documented), chat with the GitHub connector described at the bottom.

Plan

  • Write PyAutoScientist/AI_OPTIONS.md: Claude Code, Codex, free options (Copilot CLI incl. Student plan; OpenCode with limited-time Zen free models, vision caveat), "try it, not sustained science" once, Colab = see-the-code route, chat-with-GitHub-connector section at the bottom (Claude.ai Free has the connector, ChatGPT's is paid; reads the repo well, cannot run code). Fold the still-useful content of autolens's docs/setup/{claude_code,codex_cli,opencode_cli,troubleshooting}.md into it.
  • Replace each assistant README's agent-choice paragraph, "Experimental alternatives" section and chat-retired text with one short block + link; add an AI_OPTIONS.md stub beside AI_POLICY.md.
  • Retire legacy: OpenCode setup pages, docs/evaluation/agent_evaluation.md, .gemini/settings.json stubs and their AGENTS.md mention, docs/archive/ chat pages, autofit's "Colab + Gemini" notebook/script/section; update maintainer-mode and troubleshooting references; AGENTS.md "other harness" wording names Copilot CLI and OpenCode.
  • Point the pyautolabs.github.io "Ask an AI assistant" card and the three workspace READMEs' assistant paragraphs at the new page.
  • autolens_assistant edits are deferred until PR docs(colab): image-first COSMOS-Web Ring notebook, no inference (#160) autolens_assistant#161 (cosmos-web-ring-notebook) merges and releases its claim; they land on this task's branch afterwards.

Tier: glance — merge mode: auto-merge on green if the Witness passes, post-merge summary

Detailed implementation plan

Affected Repositories

  • PyAutoScientist (primary)
  • autogalaxy_assistant, autofit_assistant, autocti_assistant
  • autolens_assistant (deferred, claimed by cosmos-web-ring-notebook / PR #161)
  • pyautolabs.github.io
  • autolens_workspace, autogalaxy_workspace, autofit_workspace (README paragraph only)

Branch Survey

Repository Current Branch Dirty?
organs/PyAutoScientist main clean (unregistered worktree feature/pyautodna-stack-management, unrelated)
galaxy/autogalaxy_assistant main clean
fit/autofit_assistant main 1 untracked script (unrelated)
cti/autocti_assistant main clean
lens/autolens_assistant main 1 untracked png; claimed by cosmos-web-ring-notebook
pyautolabs.github.io main clean
lens/autolens_workspace main 1 untracked dataset.fits (unrelated)
galaxy/autogalaxy_workspace main clean
fit/autofit_workspace main clean

Suggested branch: feature/assistant-ai-options-page

Implementation Steps

  1. PyAutoScientist: add AI_OPTIONS.md (sections: Which AI runs the assistant · Claude Code · Codex · Free options: GitHub Copilot CLI, OpenCode · Troubleshooting · Colab notebooks vs the assistant · Chat instead of an agent). Link it from README.md next to the AI policy link. Source material: autolens_assistant/docs/setup/*.md (read from main), docs/archive/chat/claude_chat_paid.md + chatgpt_paid_connector.md for the chat facts, the 2026-10-23 review for free-tier facts (cite dates; free tiers churn).
  2. autogalaxy_assistant, autofit_assistant, autocti_assistant: in README.md replace lines "Choose Claude Code or Codex ... Check the provider's current plans" + "If you cannot use either agent, OpenCode ..." + the "Experimental alternatives" section with: one paragraph (runs inside an AI coding agent; Claude Code and Codex recommended; free options and chat on the PyAutoLabs AI options page → link), the clone/claude/codex snippet, the starting prompt. Add AI_OPTIONS.md stub (copy AI_POLICY.md pattern). autofit: delete ## Experimental: Google Colab + Gemini, notebooks/experimental_colab.ipynb, scripts/experimental_colab.py, docs/setup/opencode_cli.md, docs/archive/; point docs/setup/claude_code.md/codex_cli.md at the shared page or delete if fully folded. autogalaxy: delete docs/archive/README_ai_chat_assistant_section.md. All three: delete .gemini/settings.json, drop the .gemini/settings.json sentence in AGENTS.md line ~6, reword the "On any other harness (OpenCode, Gemini CLI, ...)" sentence to name Copilot CLI and OpenCode; update modes/maintainer.md chat/OpenCode references.
  3. pyautolabs.github.io index.html ~lines 246-260: "Ask an AI assistant" card links to the new page; mention Claude Code, Codex and the free options in one sentence.
  4. Workspace READMEs (autolens_workspace L26-28, autogalaxy_workspace L24-32, autofit_workspace L48 context): collapse the "Claude Code or Codex ... paid access" paragraph to one sentence + link.
  5. Deferred (after #161 merges): autolens_assistant README three-step block → shared block; move the Colab badge paragraph under "Using PyAutoLens Assistant" as the learning route; delete docs/setup/opencode_cli.md, docs/evaluation/agent_evaluation.md, docs/archive/, .gemini/settings.json; same AGENTS.md/maintainer/troubleshooting edits; keep docs/colab/.
  6. Keep llms.txt everywhere (agent routing front door, not chat legacy).
  7. Ship: ship_workspace per repo (docs only, no library-first gate); each assistant's wiki-currency/boundary CI must stay green; site builds.

Key Files

  • organs/PyAutoScientist/AI_OPTIONS.md — new canonical page
  • */AI_POLICY.md — stub pattern to mirror for AI_OPTIONS.md
  • */README.md, */AGENTS.md (line ~6 and the "other harness" sentence), */modes/maintainer.md
  • fit/autofit_assistant/{notebooks/experimental_colab.ipynb,scripts/experimental_colab.py} — delete
  • pyautolabs.github.io/index.html — assistant card

Original Prompt

Click to expand starting prompt

One PyAutoLabs-level "AI options" page for every assistant; simplify the assistant READMEs; retire chat/OpenCode/Gemini legacy

Type: docs
Target: pyautoscientist
Repos:

  • PyAutoScientist
  • autolens_assistant
  • autogalaxy_assistant
  • autofit_assistant
  • autocti_assistant
  • pyautolabs.github.io
    Difficulty: medium
    Autonomy: supervised
    Priority: high
    Filed: 2026-10-23
    Status: formalised
    Consequence: glance

@pyautoscientist @autolens_assistant @autogalaxy_assistant @autofit_assistant @autocti_assistant @pyautolabs.github.io

Request (verbatim from the human, 2026-10-23)

I want us to sort out the options recommended for users for the autolens_assistant and other assistants. Currently, we know they work brilliantly for agentic AI subs codex and claude, and the github README.md says such. We mention (hopefully near the end) that they can be used in Claude Chat of ChatGPT with subs, but this does not work well for a free option. [...] The only thing we need is a viable free agentic AI option [...] We should then make a unified page summarising all AI options for each assistant which each assistant links too, but we should prioritize this being simple, so I think 3 AI agents: claude (paid sub), codex (paid sub) and the best free options and then towards the bottom a description of the chat option with claude and chatgpt, making it clear their GitHub sync is very good. Also, in the past we introduce stuff like llms.txt which I think was to have chat support but I think we are retiring that so for each assistant check if there is legacy stuff we can retire. [...] the goal should also be to ensure the github readme is simpler.

Follow-up decisions (same day): "lets make the de facto PyAutoLabs level, as its weird for cti to link to lens"; "use opencode and copilot for free, documenting both, I agree many will have copilot access and many wont. I agree with rest."

Context (review done 2026-10-23, Fable session)

  • All four library assistants (autolens, autogalaxy, autofit, autocti) say "Choose Claude Code or Codex", retired browser-chat routes on 2026-09-10 (archived under docs/archive/chat/ in autolens + autofit), and call OpenCode an "experimental alternative — compatible, not tested". autogalaxy and autocti link to autolens's docs/setup/*.md pages: that cross-link is the thing to remove.
  • llms.txt is NOT chat legacy: it is the agent routing front door, referenced by each AGENTS.md and the wiki. Keep it.
  • Every assistant already carries a one-line AI_POLICY.md stub pointing at PyAutoScientist/AI_POLICY.md. Reuse that pattern for the new page.
  • Free landscape 2026-10-23 (sources in the Fable review): Gemini CLI free tier ended 2026-06-18 (replaced by closed-source Antigravity CLI, tiny weekly quota) so the .gemini/settings.json stubs point at a dead product; GitHub Copilot CLI is on Copilot Free (unpublished AI-credit allowance, auto model only) and free for verified students; OpenCode's client is free with 13 limited-time free Zen models (vision unclear, billing details may be needed for the key); Codex is listed on ChatGPT Free for "quick tasks" but CLI access on Free is unconfirmed. Claude.ai Free officially includes the GitHub connector (reads file contents only); ChatGPT's GitHub connector is paid-only.
  • Colab: no agentic mode. The autolens COSMOS-Web Ring notebook is a no-AI tutorial mirror of the greeting; autofit's "Colab + Gemini" notebook pastes assistant context into the sidebar by hand. A separate follow-up prompt will cover Colab Custom Instructions injection once the human's Colab test is done; this task only retires the autofit experiment and moves the autolens badge out of the getting-started block.

Scope

  1. New page PyAutoScientist/AI_OPTIONS.md (name negotiable; the one canonical page). Order: (a) Claude Code — paid; (b) Codex — paid; (c) free options: GitHub Copilot CLI (note the Copilot Student plan) and OpenCode (client free, bring-your-own model; name the Zen free models as limited-time, warn that free models often lack vision so figure checks degrade); (d) a short "Chat instead of an agent" section at the bottom: Claude.ai and ChatGPT with the GitHub connector read the assistant repo well for questions, planning and learning, but cannot run code or look at data, so they are for discussion, not fits; note Claude's connector is on Free, ChatGPT's is paid. One sentence that every free tier is "try it, not sustained science". One paragraph that Colab notebooks are the "see the code" route and the assistant the "do it in conversation" route. Keep it short: a reader should pick in under a minute. Move the useful content of autolens's docs/setup/{claude_code,codex_cli,opencode_cli,troubleshooting}.md into (or under) this page so every assistant links to one place; starting prompts stay per-assistant.
  2. Each assistant README (four repos): replace the per-repo "Choose Claude Code or Codex" paragraph, the "Experimental alternatives" section and the chat-retired text with one short block: "Runs inside an AI coding agent. Claude Code and Codex are recommended; free options and chat use are on the PyAutoLabs AI options page" + link + the three-line clone/claude/codex snippet + the starting prompt. Add an AI_OPTIONS.md stub next to AI_POLICY.md mirroring its pattern. Replace cross-links from autogalaxy/autocti into autolens's docs/setup with the new page. autofit: delete the "Experimental: Google Colab + Gemini" section, notebooks/experimental_colab.ipynb and scripts/experimental_colab.py. autolens: move the Colab badge out of the three-step getting-started block into the "learning PyAutoLens" context (keep the notebook).
  3. Legacy retirement (all four): delete docs/setup/opencode_cli.md (autolens, autofit) once folded into the page; delete docs/evaluation/agent_evaluation.md (autolens) or reduce it to the evaluation protocol only; delete .gemini/settings.json and the AGENTS.md line that mentions it; after folding the still-true chat facts into the page, delete docs/archive/ in autolens + autofit (the git history keeps it) and docs/archive/README_ai_chat_assistant_section.md in autogalaxy; update modes/maintainer.md and docs/setup/troubleshooting.md references; update AGENTS.md wording "On any other harness (OpenCode, Gemini CLI, ...)" to name Copilot CLI and OpenCode.
  4. pyautolabs.github.io: point the "Ask an AI assistant" card at the new page.
  5. Workspace READMEs (autolens/autogalaxy/autofit workspace) carry a copy of the "Claude Code or Codex ... paid access" paragraph: replace with one sentence + link. HowTo* READMEs' "Study with the assistant" blocks stay as they are.

Out of scope

  • Colab Custom Instructions injection into generated notebooks (follow-up prompt after the human's Colab test).
  • euclid_assistant (no AI-option story).
  • Validating Copilot CLI or OpenCode against the assistant benchmarks; the page says they are supported routes, documented not benchmarked.

Done when

  • One page, four READMEs each shorter than today and linking to it, no assistant links into another assistant's docs, listed legacy removed, site card updated, PRs open (library-first does not apply; all repos are docs).

🤖 Generated with Claude Code

Activity

  1. Jammy2211 commented on Oct 23, 2026

    @Jammy2211
    ContributorAuthor

    Shipped as eight PRs (awaiting human /prm; merge PyAutoScientist#88 first)

    Deferred: autolens_assistant edits (README three-step block, Colab badge move, docs/setup + docs/archive + agent_evaluation.md + .gemini removal) wait for autolens_assistant#161 to merge and release its claim.

    Heart at ship time was YELLOW for two unrelated reasons (manifest drift mismatch; stale PyAutoLens release rehearsal), recorded verbatim in each PR body. Decision-taken: do not auto-merge.

    Follow-up filed as a Mind draft (gated on a human Colab test): draft/docs/pyautohands/colab_custom_instructions_assistant_context.md.

  2. added a commit that references this issue on Oct 23, 2026
  3. Jammy2211 commented on Oct 23, 2026

    @Jammy2211
    ContributorAuthor

    Shipped and merged (2026-10-23)

    All eight PRs merged via /prm: PyAutoScientist#88, autogalaxy_assistant#36, autofit_assistant#57, autocti_assistant#36, pyautolabs.github.io#37, autolens_workspace#597, autogalaxy_workspace#260, autofit_workspace#175. autogalaxy_assistant#36 first read red on wiki-currency (baseline pinned 2026.10.18.1 vs the 2026.10.23.1 release landing mid-flight); merging main's release baseline into the branch turned it green.

    Record: PyAutoMind complete/2026/10/assistant-ai-options-page.md. Remainder re-filed: draft/docs/autolens_assistant/ai_options_page_autolens_leg.md (starts after autolens_assistant#161 merges). Colab follow-up: draft/docs/pyautohands/colab_custom_instructions_assistant_context.md.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions