agentsclimarketplace

Bridge onboard

Skill bks-lab/open-bridge/skills/bridge-onboard

Your AI coding agent starts every session knowing your repos, your clients, and how you work — a plain git repo of markdown + YAML it reads at session start, independent of model or frontend. Context compounds instead of restarting. MIT.

Install
npx -y skills add bks-lab/open-bridge --skill bridge-onboard

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 8 stars8 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

New user onboarding and reconfiguration — discovery-driven setup with permission-gated system scan, evidence-based feature suggestions, and a read-only catalogue of what else Bridge can do. Six-phase wizard (Identity / Discovery / Suggestions / Quick-Wins / Catalog / Validate) with re-entry modes for targeted later activation. Works without GitHub. Upstreams stay empty by default and wire later when the OSS upstream or your own upstream is live. Trigger: "/bridge-onboard", "onboard", "setup", "configure bridge", "new user", "ecosystem scan", "set up bridge", "reconfigure", "reconfigure bridge", "setup wizard".

SKILL.md

10.0 KB, as published. Nobody here has run it

Bridge Onboard — Setup Wizard

Set up a new Bridge user or reconfigure an existing one. Read the referenced file ONLY when triggered.

Philosophy — Discover, don't Interrogate

A new user came here to configure an assistant, not to inventory their life. Asking abstract questions like "do you file taxes for multiple legal entities?" or "do you have a household mandant?" makes the user defensive — they don't know yet what those features do, so they say "skip" to everything and miss the point.

Instead, the wizard:

  1. Looks at what's already on the system (with permission)
  2. Proposes specific features that match the evidence
  3. Shows the rest as a read-only catalogue with "when you need it"
  4. Surfaces features proactively later (feature-discovery standing-order) when patterns suggest they'd help

All of this is consent-first: before step 1, the user picks discovery.modeconfined (default) or broader. Confined means the Bridge never scans the machine (other repos, installed apps, devices, files, and mail stay untouched) — you still get every feature, just enabled modularly when you want it via the Phase E catalog, --add <feature>, or the feature's enabled: flag in bridge-config.yaml. Only broader unlocks the permission-gated scan in step 1.

The entry point is the four-lane front door in rules/session-start.md § NEW USER front door — it reflects what Phase 0 detected (origin state, git name), notes the guard is already armed, and opens four discoverable lanes under a free-text invite: [1] show me around (demo) · [2] describe what I'll use it for (→ this wizard, purpose pre-filled) · [3] make it private first (protection) · [4] bind a workspace / org overlay. This file and references/ are where lanes [2]/[3]/[4] land; the heavy lifting lives in references/.

Defaults — what the wizard assumes

  • Purpose pins a north-star (ordering, never gating). It is captured at the front door — the user describes what they're here to do (session-start lane [2] or a free-text answer) and that sentence becomes purpose.statement verbatim; Phase A does not re-ask. purpose.focus and user_profile are derived silently from the statement — never posed as a visible six-domain questionnaire. Purpose ORDERS the Phase C suggestions, the Phase E catalogue, and (under broader) the feature-discovery standing-order; it never hides, gates, or removes a feature. Empty purpose = today's flat, general-purpose behaviour. Change anytime via --purpose.
  • GitHub is optional. Onboarding completes end-to-end without a GitHub org, without gh CLI, and without GitHub-projects integration.
  • Upstreams stay empty (upstreams: []) by default. Upstream variant choice is a separate, optional step via --upstream.
  • Work-system on by default. Phase D recommends enabling — it's what makes Claude resume context across sessions. The wizard explains the trade-off but suggests [y].
  • Discovery scan is opt-in per source. Default-on sources are non-invasive (git config, dir listings, app list). Sensitive sources (mail accounts, finance-app accounts, calendar names) are default-off and only opt-in.
  • Nothing scanned beyond names. Mail content, document content, message bodies, keychain, passwords — NEVER touched. Explicit in the permission prompt so trust is established.

Privacy & Trust

Scan findings live in work/onboarding-scan.json (gitignored, auto-deleted after 30 days or on --reset). Granted permissions are persisted to bridge-config.yaml.discovery.permissions so re-runs honor the same boundary without re-asking.

Decision history lives in work/onboarding-state.yaml (NOT gitignored — useful as a record of setup choices) with statuses accepted | deferred | declined | silenced | nothing_found.

The feature-discovery standing-order (active when work.enabled: true) uses the same state file to avoid double-suggesting.

Mirror-safety — check the origin before writing any private data. Onboarding writes the user's identity and work/ to a user/{name} branch. Before creating that branch (Phase A step 6), resolve where the clone pushes: git remote get-url origin (and gh repo view --json visibility,nameWithOwner if unsure). If origin is a PUBLIC repo or a known upstream (e.g. bks-lab/open-bridge) — or .bridge-origin says is_public: true — STOP and advise, do not proceed silently. The user's private data must not live on a public origin. Offer to set them up on their own private repo first (GitHub Use this template → Private, or re-home origin to a new private repo with open-bridge as a read-only upstream), then continue. Never push the user/* branch to a public origin; CORE reaches a public upstream only via /promote. Canonical rule: ../../rules/push-guard.md.

Modes

InvocationBehaviour
/bridge-onboardFull wizard (Phases A–F)
/bridge-onboard --rescanBroaden discovery (sets discovery.mode: broader); re-run Phase B+C with persisted permissions; surface new evidence; skip already-accepted features
/bridge-onboard --resetDelete scan + state files; restart from Phase A (re-runs the scope-consent gate + purpose). Prompts to delete bridge-config.yaml for true clean-slate
/bridge-onboard --add <feature>Skip A+B+D+E+F, run only the matching S-block from smart-suggestions.md (e.g. --add personas, --add doc-system)
/bridge-onboard --add agent-soulSkip everything except D4 — re-pick the soul deck and reshape SOUL.md / IDENTITY.md
/bridge-onboard --purposeSkip everything except the Phase-A purpose step — set/change purpose.statement + purpose.focus (re-derive user_profile), then re-render the Phase F preview ordering. Never gates a feature
/bridge-onboard --featuresRead-only Phase E catalogue, interactive — explore what Bridge can do, click into entries to activate
/bridge-onboard --upstreamSkip everything except upstream wiring (see below)

Optional: --upstream mode

When invoked as /bridge-onboard --upstream, the wizard skips Phases A-E, going straight to upstream configuration. Use cases:

  • User already onboarded, now wants to wire bks-lab/open-bridge once it is publicly available
  • Forked the Bridge into a private org and wants to push contributions

In this mode, ask:

Which upstream are you wiring?

  [1] open-bridge (OSS) — sets contribute: true, pull-only OSS-core
  [2] org-internal bridge — sets primary: true, push-enabled
  [3] Custom — type repo/branch/role manually

Each choice appends one entry to upstreams: [] and adds a matching promote.content_blocklist.<name> skeleton (user fills the strings/patterns based on what their codebase shouldn't leak).

Decision Tree

User wants to...
├── Full onboarding wizard             → Read references/workflow.md
├── Re-scan and resurface deferred     → Read references/system-discovery.md (--rescan)
├── Start fresh                        → Read references/system-discovery.md § Re-Run Modes (--reset)
├── Activate one specific feature      → Read references/smart-suggestions.md § Adding a Feature Later
├── Set or change the instance purpose → Read references/workflow.md § Phase A step 5 (--purpose)
├── Browse what Bridge can do          → Read references/feature-catalog.md (--features)
├── Wire an upstream after the fact    → This file § Optional: --upstream mode
└── Questions about setup              → Answer from references/workflow.md + CLAUDE.md § Session Start

Reference Files

FilePurpose
references/workflow.mdSix-phase wizard execution plan (entry point for full onboarding)
references/system-discovery.mdPhase B — what gets scanned, with which permission, what's never scanned
references/smart-suggestions.mdPhase C — evidence → recommendation mapping (S1–S14) with full advisory text
references/feature-catalog.mdPhase E + --features — read-only catalogue of all Bridge features
references/discovery.mdLegacy: repo-only ecosystem detection. Now a sub-case of system-discovery.md
references/preview-generator.mdPhase F — HTML preview with Activated + Suggested-for-later sections

Related Files

  • protocols/standing-orders/feature-discovery.md — proactive feature suggestions in the weekly briefing window (uses work/onboarding-state.yaml to avoid double-suggesting)
  • docs/feature-tour.md — per-cluster-wrapper "create your first X" guide for after onboarding
  • bridge-config.yaml.template — every block has a comment pointing at the matching --add <feature> re-entry path

Implementation Notes

  • All wizard output is bilingual where the user's language is German (matches language.conversation from Phase A)
  • Suggestions in Phase C run sequentially, not as a batch — each [y] scaffolds immediately so the user sees progress
  • Errors during scan or scaffold are non-fatal; record and continue, surface the issue in Phase F's validation step
  • Never write to bridge-config.yaml outside of explicit phases; the user must always see the current state before changes are committed

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.