takazudo/dev-setup-temp-resource

Dev Setup: Temp Resource

A repo-committed handoff channel for scratch resources that one session produces and a later session consumes. Git is the channel — so it works where Dropbox/cclogs does not (most importantly Claude Code web, which has no Dropbox).

The convention

• Location: _temp-resource/ at the repo root, committed (NOT gitignored). Being in git is the whole point — the resources ride the branch/PR to whatever machine or web session picks the work up.
• Per topic: _temp-resource/<issue-number>-<topic-slug>/, e.g. _temp-resource/4444-tweak-header/. The number is the GitHub issue — the epic for /big-plan, the single issue for /x-as-pr.
• Temporary: delete the topic subdir when the delegated work merges, so it never reaches the default branch. Harmless if left (tooling ignores the dir), but prefer clean.

When to use it — and when not

Use it only when delegating to a later session AND the resource is not already in the repo and is too big/binary to express inline in the issue (a prototype, a design mockup + screenshots, test fixtures, sample data).

Do not reflexively store everything. /prototype-first-wisdom output is often discarded or never acted on — keep that in cclogs. Promote a copy into _temp-resource/ only once you've decided a downstream session concretely needs it. In-session work needs none of this; just use the file directly.

One-time repo setup

A stray prototype (.ts/.tsx/.md/.html) under _temp-resource/ must not trip CI's format / lint / test gates. Most repo tooling is path-scoped (e.g. pnpm --filter ...) and never sees a root dir — but the broad scanners (markdown formatters, repo-wide linters, root test configs) will, so they need an explicit exclude. Run the helper:

bash "$HOME/.claude/skills/dev-setup-temp-resource/scripts/ensure-temp-resource.sh"

It creates _temp-resource/ + a README.md, verifies the dir is not gitignored, and auto-patches the configs whose shape is known and safe to edit:

• file-based ignore lists — .prettierignore / .eslintignore / .stylelintignore.
• format — .mdx-formatter.json → adds _temp-resource/** to its exclude array.
• lint — .design-token-lint.json → adds **/_temp-resource/** to its ignore array.

Then it reports the remaining lint / test / typecheck configs for you to finish by hand (their shapes vary too much to patch blindly):

• lint — eslint.config.* (flat) → "_temp-resource/**" in an ignores entry; .eslintrc* → ignorePatterns; biome.json → files.ignore.
• test — vitest.config.* / vite.config.* (if it defines test) → "_temp-resource/**" in test.exclude; jest.config.* → testPathIgnorePatterns; playwright.config.* is usually testDir-scoped (safe — just confirm testDir isn't the repo root).
• typecheck — tsconfig*.json → "_temp-resource" in exclude, but only if its include globs the repo root (per-package tsconfigs that scope to their own src need nothing).

Edit the tool's own config array — never add a CLI --ignore/--exclude flag to the npm script. For tools that auto-load a config (mdx-formatter's .mdx-formatter.json, etc.), a CLI flag replaces the config's list instead of extending it, silently re-exposing everything else the config was excluding (e.g. generated docs). The helper edits the config array for exactly this reason.

Commit this setup once (its own small PR, or fold it into the first delegating base branch). It is permanent repo plumbing; the per-topic subdirs come and go on feature branches.

Store a resource (the delegating session)

1. Pick the issue number + a short topic slug.
2. mkdir -p _temp-resource/<issue>-<slug>/ and put the files there.
3. In the issue body, reference them by this in-repo path — never a $DROPBOX_*/$HOME/cclogs/absolute path (those don't exist on another machine or on web). A compact inline summary (e.g. ASCII of a layout) alongside the path makes the issue self-contained even before the file is opened.

The base-branch delegation protocol — /big-plan → /x-wt-teams

By default /x-wt-teams creates the base branch and /big-plan creates issues only. When resources must be delegated, /big-plan prepares the base branch instead, and /x-wt-teams reuses it:

1. /big-plan (after the epic + sub-issues exist, so the epic number is known):

• Create base/<impl-title-slug> from the parent branch.
• Commit the resources under _temp-resource/<epic#>-<slug>/; push; open the base PR (base/<slug> → parent).
• Add to the epic body: a "Use this PR as base" note naming the base branch + base PR and the _temp-resource/<epic#>-<slug>/ path. This is the cross-session message.

2. /x-wt-teams: when the epic says "use this PR as base", reuse that existing base branch — do NOT create a new one. Topic branches fork from it, so every child inherits the resources, and each sub-issue can point at _temp-resource/<epic#>-<slug>/ in the working tree.

For Claude Code web this is the only reliable handoff: the PR + committed dir ARE the shared state.

Consume (the later session / child agents)

The resources are already on the branch — read _temp-resource/<issue>-<topic>/ straight from the working tree. The issue body points there. No download step, no Dropbox.

Cleanup

When the delegated work merges, remove _temp-resource/<issue>-<topic>/ so it doesn't land on the default branch — fold this into the consuming workflow's cleanup/confirm step (or the docs step that migrates anything durable into real docs). Leaving it is harmless but untidy.