Manage · Repo setup prompt
Review a full repository setup before an agent writes it
The Repo setup prompt is a broad recipe drawn from this repository. Read it as guidance for an agent, not as an installer or proof that setup worked. Start with the reviewed deterministic setup when you need the initial brain.
For: a repository maintainer deciding which memory, skill, documentation, workflow, and CI conventions belong in a project
What it does, and why it helps
The desktop tab renders .claude/brain/repo-bootstrap-prompt.md from the selected repository. The RepoOps copy asks an agent to inspect the project, confirm scope, work on a branch, build a broad file set, and open a pull request. Its inventory spans root instructions, brain files, skills, scripts, documentation, CI, and GitHub files.
The page only renders markdown. It does not run the prompt, inspect the target, preview writes, constrain the agent, or verify the result. The prompt is specific to RepoOps and its current working rules. For an initial brain, the reviewed repository setup flow produces a bounded source-backed plan and preserves existing instructions. The public generic prompt is a separate file with a version header.
The pain. A mature repository may need more than four memory files, but copying a whole project structure can also import CI jobs, permissions, skills, and workflow rules that do not fit the target.
The point of view. Use deterministic setup for facts and a prompt for judgment. A broad recipe is useful when every proposed file is reviewed against the target repository, not when the recipe is treated as a universal install script.
What gets easier. The prompt keeps the RepoOps repository setup inventory in one readable source. It marks several artifacts as conditional and gives a reviewer one checklist for the proposed branch and pull request.
When it helps. Use it after the initial brain exists, when a maintainer wants to evaluate the larger RepoOps process layer: skills, docs structure, CI gates, generated-view rules, and the completion ritual.
Its limits. It is a legacy AI-assisted workflow. No code interprets or enforces its steps. The prompt can become stale when the repository changes, and its broad instructions can produce a large diff. Creating brain files does not activate assistant capture or team delivery.
Understand it in 30 seconds
Read the narration
- 0:00 A broad setup prompt can turn one generic recipe into a repo-wide diff.
- 0:06 Start with deterministic setup. Use the prompt only for choices that need judgment.
- 0:13 The prompt asks an agent to inspect the stack and mark conditional files.
- 0:17 Then it asks for a branch and a reviewable pull request.
- 0:23 Review the diff. A prompt is guidance, not proof that setup worked.
Synthetic example. Read the guide
Where to find it
Where to find it
- Desktop:
localhost:4000, then Local settings in the sidebar, then Repo setup prompt under Workspace utilities. - Keyboard: ⌘ K, then type “Repo setup prompt”.
When to use it
Evaluate the full process layer for an established repository
Situation. A repository already has its own commands and contribution rules, but the team wants to compare them with the RepoOps brain, skills, docs, CI, and completion ritual.
What you do. Read Repo setup prompt from the selected RepoOps repository. Before pasting, remove sections that do not apply and state which existing files must be preserved. Use a fresh agent session on its own branch or worktree, and ask for scope confirmation before writes.
What you see. The proposed diff names each new or changed file. Conditional items appear only when the target has the matching need. Existing commands, provider config, and instruction text remain grounded in the target rather than copied from RepoOps.
What it establishes. You have a reviewable proposal for a larger process layer. The proposal does not establish that its commands pass, its CI gates fit, or its agent permissions are safe until you check them.
Add judgment after deterministic memory setup
Situation. The reviewed repository setup has created a minimal source-backed brain, and the project now needs selected skills or workflow rules that cannot be inferred from manifests alone.
What you do. Compare the full prompt with the applied brain-only plan. Select the few additional artifacts the project needs, such as a docs skill for a user-facing docs site or a website-build job for a compiled site. Leave unrelated sections out.
What you see. The pull request separates deterministic brain files from judgment-heavy additions. Repository checks exercise the added scripts and jobs, and the review explains every conditional inclusion.
What it establishes. The project gains only the reviewed additions. Assistant capture and hosted delivery remain separate setup tasks with their own evidence.
Before you start
- Supported versions
- RepoOps desktop v0.3.1, the release this guide was read against. The rendered prompt has no version header; the separate public generic prompt identifies Prompt version 1.0.0 and RepoOps kernel v1.0.0.
- Where it runs
- Shared scope. Desktop: Local settings, Workspace utilities, Repo setup prompt for the selected repository. The command palette can open the same leaf. Hosted: Help links to a bundled Repo setup prompt in the shared document viewer; a permitted published repository copy can take precedence when present.
- Permissions
- Reading the local page needs access to the tracked repository. Reading the hosted viewer needs a signed-in team membership and applies repository scope to snapshot reads. Running the text is outside this feature and inherits the selected agent tool, filesystem, git, provider, and repository permissions.
- Connections
- Local rendering needs no network or model connection. Hosted rendering needs the RepoOps team dashboard. Pasting the prompt needs whichever coding agent and provider the operator chooses. Initial reviewed setup uses the local app or authenticated CLI instead.
- Plan
- No prompt-specific plan gate exists. The bundled hosted document follows team dashboard access. Agent-provider charges and repository permissions come from the external tool, not from this prompt surface.
Configure it
- Begin with reviewed deterministic setup.
Use the repository setup flow for the initial source-backed brain. It inspects without running repository scripts, shows exact file changes, detects intervening edits through a fingerprint, and preserves existing instructions. The full prompt is optional afterward.
- Read the source that matches your intent.
The Repo setup prompt in the dashboard is the internal RepoOps recipe. The public generic prompt printed by init --prompt is a separate versioned artifact. Do not treat the internal inventory as a portable default for every project.
- Cut the prompt to the target scope.
Keep only files and gates supported by the target repository. State existing-file preservation rules and ask the agent to report inapplicable sections before it writes.
- Isolate the write and review the diff.
The prompt asks for a branch and pull request, but no RepoOps module enforces that request. Use the repository workflow to isolate the agent and inspect every file before merge.
- Verify activation separately.
Run the target repository checks. Then verify assistant capture and team delivery through their own setup flows. A new brain directory or merged prompt-generated pull request proves neither connection.
| Setting | Where | A sensible choice | Why it matters |
|---|---|---|---|
.claude/brain/repo-bootstrap-prompt.md | Selected repository, rendered by Repo setup prompt | Keep the RepoOps copy as reference; edit a reviewed target-specific copy when the policy should differ | The page reads this markdown at request time. Editing it changes future guidance but does not update files previously produced from an older copy. |
.claude/settings.json | Prompt file inventory | Review every proposed permission and hook against the target repository | The prompt proposes acceptEdits, an allowlist, and hooks. These are instructions to an agent, not a permission policy generated or checked by the prompt page. |
AGENTS.md | Prompt root inventory | Include only when framework or cross-tool instructions need it | The prompt marks it conditional. Existing instruction text must be preserved and reconciled rather than replaced with RepoOps-specific rules. |
documentation/ | Prompt skills inventory | Include only for a repository with a user-facing docs site | The old page and prompt both treat the documentation skill as conditional. A library or service with no docs surface does not gain value from a copied docs workflow. |
website-build | Prompt CI inventory | Include only when the repository builds a compiled site | The prompt marks this CI job conditional. Its command must come from the target manifest and pass there before the job is accepted. |
autoresearch/ | Prompt skills inventory | Omit unless the target has a reviewed use for the optional research workflow | The prompt labels this skill optional. There is no feature control that decides applicability or configures its provider access. |
remotion/ | Prompt skills inventory | Include only when the repository renders videos | The prompt marks it conditional on rendered video work. Adding a directory does not install or configure a rendering toolchain. |
What you should see
Prompt renders from the selected repository
Configuration. The tracked repository contains .claude/brain/repo-bootstrap-prompt.md and the Repo setup prompt leaf is opened.
Expect. RepoOps renders the markdown in the shared prose shell. No committed HTML twin is required. The shell title reads Repo setup prompt, and the source H1 is Repo bootstrap prompt, the full gold-standard setup.
Verify. Confirm the footer names repo-bootstrap-prompt.md as the rendered source and compare the visible content with the file in the selected repository.
Missing source stays explicit
Configuration. The selected repository has no repo-bootstrap-prompt.md under its brain directory.
Expect. The registered leaf renders its zero state instead of returning a silent blank page. It says that the full repo setup prompt is not present and names the brain file needed to fill the page.
Verify. Read Nothing here yet and the next action. Do not infer that the repository has already received any bootstrap files.
Reviewed agent proposal
Configuration. A tailored copy is pasted into an external coding agent on an isolated target branch, with conditional sections removed first.
Expect. If the agent follows the request, it inspects the target, reports scope, proposes repository-specific files, and opens a reviewable change. This is an expectation from prose, not an enforced state machine.
Verify. Inspect the actual diff, run the target checks, compare each new rule with the repository, and verify capture and team delivery separately. Stop if the agent copied RepoOps commands or policy without target evidence.
Data and cost
- What is captured
- The feature captures no run, acceptance, output, or agent transcript. Its source is markdown in the repository. The hosted build also carries a generated bundled copy of the four bootstrap documents, including repo-bootstrap-prompt.md.
- Who can see it
- Anyone with read access to the repository can read its prompt. A signed-in team member can open the hosted document viewer. When an allowed brain snapshot contains the same file key, that repository copy is shown; otherwise the viewer uses the bundled RepoOps copy.
- How long it is kept
- No prompt-specific retention setting exists. The local source follows repository history. The bundled copy follows product releases. A published copy follows the team brain snapshot store. The prompt page does not remove any of them.
- What leaves the machine
- Local rendering makes no model call and sends no prompt. Hosted viewing requests the bundled or team-scoped document from RepoOps. Pasting the text into a coding agent can send the prompt and whatever repository context that tool reads to its provider; this feature does not inspect or limit that transfer.
- What it costs
- Viewing and editing the markdown has no model charge. Any agent run uses the selected tool and provider account. The prompt sets no token cap, model, budget, or price, so no fixed cost can be stated.
When the result differs
| Symptom | Likely cause | Next action |
|---|---|---|
| The page says Nothing here yet. | The selected tracked repository does not contain .claude/brain/repo-bootstrap-prompt.md. | Use the deterministic repository setup for the initial brain. Add a reviewed prompt source only if this legacy workflow is still needed. |
| You expected Copy or Run, but the page only shows prose. | The leaf is a markdown viewer. No public static page or client script adds a copy or execution control. | Select text manually if you intend to use it, or choose the reviewed repository setup flow when you need a preview and apply action. |
| The agent proposes a repo-wide set of unrelated files. | The internal prompt describes the full RepoOps setup and asks for brain, skills, scripts, docs, CI, and GitHub machinery together. | Stop before accepting writes. Remove inapplicable sections, name the allowed paths, and request a smaller plan grounded in the target repository. |
| The agent adds generated brain HTML twins. | The file inventory still names historical dashboard HTML artifacts, while the prompt later says generated prose twins are retired and must not be tracked. | Follow the current HTML twins section. Keep markdown as source and include only a hand-curated bespoke HTML page when the target truly has one. |
| The proposed settings file grants more agent access than expected. | The prompt proposes acceptEdits and a generous allowlist but cannot inspect or enforce the target security policy. | Reject that part of the diff. Review each command and hook under the target tool security model before adding it. |
| The new brain exists, but no assistant session or hosted evidence appears. | Prompt-generated repository files do not activate assistant capture, device binding, publishers, or team delivery. | Use the repository connection flow, then verify a captured session and a hosted receipt through their own surfaces. |
| Hosted and desktop copies differ. | The desktop reads the selected repository file. The hosted viewer prefers an allowed published snapshot and falls back to the generated bundled product copy when no snapshot row is found. | Read the source line shown by each viewer. Update and publish the repository copy if that is the intended team version, or use the bundled copy as the product reference. |
- Disable
- Not provided. The prompt has no scheduled job, hook, or execution switch. Do not paste it, or stop the external agent before accepting writes.
- Roll back
- Not provided. The prompt does not record a write manifest or checkpoint. Review and reverse its changes through the target repository version-control workflow. The deterministic setup flow has its own maintenance plan and is separate from this feature.
- Revoke access
- Not provided. Reading the prompt creates no token or access grant. Revoke any coding-agent, provider, repository, or team access in the system that issued it, outside this feature.
- Delete
- Not provided as one action. Delete or revert prompt-generated files through the target repository after reviewing ownership and dependencies. Removing the local markdown does not remove the bundled hosted product copy.
Related tasks
Maintenance evidence
- Feature id
repo-bootstrap(spine leafrepo-bootstrap-prompt)- Owner
- Gold-standard infrastructure program. Guide: LDG-0717.
- Supported product version
- RepoOps v0.3.1
- Last verified
- 2026-09-15, read against origin/main at 52366bb6d; labels read from the served tab source; canonical registry, internal prompt, markdown renderer, deterministic setup, public prompt, hosted bundle, hosted viewer, and tests checked in the repository.
- Example fixtures
- No prompt-run fixture exists because this feature does not execute the text. lib/routes/mirror-files.twin-fallback.test.mjs and lib/md-renderer.frontmatter.test.mjs cover markdown fallback and frontmatter. scripts/gen-bootstrap-bundle.test.mjs pins the hosted bundle to the four source documents. lib/install/scaffold-into-repo.test.mjs covers the reviewed brain-only alternative, exact preview and apply parity, existing-file preservation, manifest bounds, and non-git refusal. scripts/kernel-init.test.mjs covers the compatibility stamp, idempotence, preservation, force overwrite, and kernel gates. bin/repoops-init-birth.test.mjs covers the versioned public prompt and kernel command.
- Source references
lib/canonical-spine.json,lib/canonical-tabs.mjs,.claude/brain/repo-bootstrap-prompt.md,lib/routes/mirror-files.mjs,lib/md-renderer.mjs,lib/zero-states.mjs,lib/install/scaffold-into-repo.mjs,lib/install/repository-integration.mjs,scripts/kernel-init.mjs,docs/distribution/repo-brain-bootstrap-prompt.md,bin/repoops-init.mjs,scripts/gen-bootstrap-bundle.mjs,website/lib/bootstrap-docs-content.ts,website/lib/dashboard-nav.ts,website/app/team/(home)/docs/[file]/page.tsx- Documentation review
- Independent review requested on the slice pull request; not yet recorded.
- Video review
- Narrated story rendered and published 2026-09-26 (render 9b1a983cae16, LDG-1018) with the breadcrumb Local settings, checked against main at 0f24215b5. One frame, the captions and the transcript were reviewed by the authoring agent, not an independent reviewer; the audio was not listened to by a person. Narration is the provisional Windows voice until LDG-0721.
Last updated