Add research-agent-skills

[jjans5]

Adds research-agent-skills to the marketplace: three on-demand skills (no hooks) for research projects where agents write code, run compute and hand work over.

Skill	What it does
analysis-code	Writing and changing research code: one owner per operation, small pure functions, correctness traps (silent drops in joins, pooled metrics, ignored seeds), verify before reporting.
research-workflow	Scope and stop rule, pre-registered PLAN.md before compute, PROGRESS.md / HANDOVER.md, briefs for subagents and Codex, runlock.py (one writer per output folder).
project-setup	One CLAUDE.md read by both Claude Code and Codex, CODE_MAP.md / ENVIRONMENTS.md drafted by survey scripts, handover/ index.

How it relates to the existing plugins: in repos that enable analysis-workflow@quadbio, all three skills defer to its conventions (AGENTS.md, task directories, worktrees, working objects), and project-setup's survey detects such repos and says so. Cluster specifics stay in euler-hpc, figures in sci-figures.

Unpinned on purpose. The entry has no ref, and the plugin's plugin.json has no version. Claude Code then versions it by commit SHA, so every merge to its main reaches users with auto-update on, with no marketplace bump needed. The plugin's README documents this. Happy to switch to tag pinning like the other plugins if you prefer that for the marketplace.

🤖 Generated with Claude Code

[Marius1311]

@codex review

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-13T12:10:34.337355Z	86284db	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Hooray!

Reviewed commit: 86284db451

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Marius1311]

Thanks @jjans5 a welcome addition. The entry itself is right as it is: #11, merged just after you opened this, moved every entry to follow its plugin's main, so no ref is now the norm. Plugins in this marketplace should work alike, so three changes in research-agent-skills before I merge:

1. Release by version bump. Add "version": "0.1.0" to .claude-plugin/plugin.json and use the Release section the other plugins have: "Bump version in .claude-plugin/plugin.json and merge: the marketplace follows main, and users update when the version changes." A merge to main then doesn't reach everyone at once; a release is a deliberate bump. Install can shrink to a pointer to this repo's README, as in euler-hpc.
2. AGENTS.md as the one instruction file. Claude Code now reads AGENTS.md itself when there is no CLAUDE.md (v2.1.277+, docs), and Codex reads it natively, so one AGENTS.md tracked in the repo serves both agents in every worktree, with no settings. analysis_template already does this. That makes project-setup's rename to CLAUDE.md (§5) and the Codex config changes and marker files (§6) unnecessary: please switch the skill, its template and setup_project.py to AGENTS.md and drop those two sections.
3. Worktrees under .claude/worktrees/, where claude --worktree <name> creates them and analysis-workflow keeps them, instead of <project>/worktrees/<date>_<slug>.

Minor: research-workflow points to a token-lean skill "if it is installed"; no plugin here ships one, so drop the mention or add that skill to the marketplace.

[jjans5]

Thanks @Marius1311, all three are done in quadbio/research-agent-skills#1, now merged (0.1.0):

1. Releases by version bump. version is in plugin.json, and Release and Install read as in the other plugins.
2. AGENTS.md as the one instruction file. project-setup, its template and setup_project.py create and keep a tracked AGENTS.md, and convert CLAUDE.md / .claude/CLAUDE.md / CLAUDE.local.md / agents.md into it. The rename-to-CLAUDE.md step and the Codex config and marker files are gone. Since a CLAUDE.md in any folder above the project also stops Claude from reading AGENTS.md, the survey reports those too.
3. Worktrees under .claude/worktrees/<name>. The survey also flags a missing .gitignore entry.

Minor: the token-lean pointer is dropped.

The entry in this PR is unchanged (no ref, following main), so it should be ready.